From ad2d8a3cf393d68209328a53c0837266b80a2628 Mon Sep 17 00:00:00 2001 From: xiaodaigh Date: Sat, 8 Aug 2020 13:01:21 +1000 Subject: [PATCH 1/8] minor --- DESCRIPTION | 2 +- R/dplyr_verbs.r | 24 ++++++++++-------------- 2 files changed, 11 insertions(+), 15 deletions(-) diff --git a/DESCRIPTION b/DESCRIPTION index 421d1903..d235b10b 100644 --- a/DESCRIPTION +++ b/DESCRIPTION @@ -52,7 +52,7 @@ Suggests: covr LinkingTo: Rcpp -RoxygenNote: 7.1.0 +RoxygenNote: 7.1.1 Encoding: UTF-8 URL: https://diskframe.com BugReports: https://github.com/xiaodaigh/disk.frame/issues diff --git a/R/dplyr_verbs.r b/R/dplyr_verbs.r index 9daca00e..8ea3c4ea 100644 --- a/R/dplyr_verbs.r +++ b/R/dplyr_verbs.r @@ -83,20 +83,16 @@ tally.disk.frame <- create_chunk_mapper(dplyr::tally) #' @rdname dplyr_verbs count.disk.frame <- create_chunk_mapper(dplyr::count) -# TODO family is not required is group-by -# TODO alot of these .disk.frame functions are not generic - - -#' @export -#' @importFrom dplyr add_count -#' @rdname dplyr_verbs -add_count.disk.frame <- create_chunk_mapper(dplyr::add_count) - - -#' @export -#' @importFrom dplyr add_tally -#' @rdname dplyr_verbs -add_tally.disk.frame <- create_chunk_mapper(dplyr::add_tally) +#' #' @export +#' #' @importFrom dplyr add_count +#' #' @rdname dplyr_verbs +#' add_count.disk.frame <- create_chunk_mapper(dplyr::add_count) +#' +#' +#' #' @export +#' #' @importFrom dplyr add_tally +#' #' @rdname dplyr_verbs +#' add_tally.disk.frame <- create_chunk_mapper(dplyr::add_tally) #' @export From 461f88608bd3ca85f5255e42c99aab656b5fee03 Mon Sep 17 00:00:00 2001 From: xiaodaigh Date: Sat, 8 Aug 2020 13:04:26 +1000 Subject: [PATCH 2/8] minor cleanup --- CRAN-RELEASE | 4 +- book/10-group-by.Rmd | 33 +- .../rstudio conf 2021/1min video script | 11 + .../rstudio conf 2021/Abstract proposal.md | 5 + presentation/twin cities/.gitignore | 1 + tests/testthat/test-add-chunk.r | 24 ++ tests/testthat/test-names.r | 7 + vignettes/concepts.Rmd | 69 ---- vignettes/convenience-features.Rmd | 54 --- vignettes/data-table-syntax.Rmd | 78 ---- vignettes/glm.Rmd | 111 ----- vignettes/ingesting-data.Rmd | 174 -------- vignettes/intro-disk-frame.Rmd | 378 ------------------ 13 files changed, 67 insertions(+), 882 deletions(-) create mode 100644 presentation/rstudio conf 2021/1min video script create mode 100644 presentation/rstudio conf 2021/Abstract proposal.md create mode 100644 presentation/twin cities/.gitignore delete mode 100644 vignettes/concepts.Rmd delete mode 100644 vignettes/convenience-features.Rmd delete mode 100644 vignettes/data-table-syntax.Rmd delete mode 100644 vignettes/glm.Rmd delete mode 100644 vignettes/ingesting-data.Rmd delete mode 100644 vignettes/intro-disk-frame.Rmd diff --git a/CRAN-RELEASE b/CRAN-RELEASE index 116e1f09..598e45ee 100644 --- a/CRAN-RELEASE +++ b/CRAN-RELEASE @@ -1,2 +1,2 @@ -This package was submitted to CRAN on 2020-02-24. -Once it is accepted, delete this file and tag the release (commit eea20eac59). +This package was submitted to CRAN on 2020-07-07. +Once it is accepted, delete this file and tag the release (commit eb830e2ac0). diff --git a/book/10-group-by.Rmd b/book/10-group-by.Rmd index f0f6ec34..1f47eb55 100644 --- a/book/10-group-by.Rmd +++ b/book/10-group-by.Rmd @@ -58,22 +58,23 @@ It is important to note that not all functions that can run in `dplyr::summarize If a function you need/like is missing, please make a feature request [here](https://github.com/xiaodaigh/disk.frame/issues). It is a limitation that function that depend on the order a column can only obtained using estimated methods. -| Function | Exact/Estimate | Notes | -| -- | -- | -- | -| `min` | Exact | | -| `max` | Exact | | -| `mean` | Exact | | -| `sum` | Exact | | -| `length` | Exact | | -| `n` | Exact | | -| `n_distinct` | Exact | | -| `sd` | Exact | | -| `var` | Exact | `var(x)` only `cor, cov` support *planned* | -| `any` | Exact | | -| `all` | Exact | | -| `median` | Estimate | | -| `quantile` | Estimate | One quantile only | -| `IQR` | Estimate | | +| Function | Exact/Estimate | Notes | +|--------------|----------------|--------------------------------------------| +| `min` | Exact | | +| `max` | Exact | | +| `mean` | Exact | | +| `sum` | Exact | | +| `length` | Exact | | +| `n` | Exact | | +| `n_distinct` | Exact | | +| `sd` | Exact | | +| `var` | Exact | `var(x)` only `cor, cov` support *planned* | +| `any` | Exact | | +| `all` | Exact | | +| `median` | Estimate | | +| `quantile` | Estimate | One quantile only | +| `IQR` | Estimate | | + ### Notes on One-Stage group-by diff --git a/presentation/rstudio conf 2021/1min video script b/presentation/rstudio conf 2021/1min video script new file mode 100644 index 00000000..53337167 --- /dev/null +++ b/presentation/rstudio conf 2021/1min video script @@ -0,0 +1,11 @@ +Hi my name is ZedJ and I am a Data Scientist local to Melbourne. I am a keen contributor to open source data science projects, one of which I want to talk about at rstudio:conf 2021. That project is {disk.frame} - a larger-than-RAM data manipulation package. + +R needs to load the data in its entirety into RAM. However, RAM is a precious resource and often do run out. + +{disk.frame} solves this issue by providing a 100%-R framework to manipulate data on disk. A modern laptop with {disk.frame} can comfortably handle 100GB's of data. + +Also, {disk.frame} uses {dplyr} verbs to manipulate data so useRs will find it very easy to pick up. + +Finally, because {disk.frame} is 100%-R, you can use any R package with it at no extra cost unlike Spark. + +The talk I propose will introduce {disk.frame} to users with the needs to manipulate large amounts of data with minimal setup. They will find {disk.frame} very familiar, as {disk.frame} uses {dplyr} verbs directly;. Some users rely on DBMS (e.g. PostgresSQL), Spark, or SAS to manage their large datasets. They will find lots of benefits in switching to {disk.frame}, which will allow them to keep their workflow in R for as long as possible. Because {disk.frame} can run R functions natively, they will find that {disk.frame} allows them to many R packages directly with {disk.frame}. diff --git a/presentation/rstudio conf 2021/Abstract proposal.md b/presentation/rstudio conf 2021/Abstract proposal.md new file mode 100644 index 00000000..e7406d3f --- /dev/null +++ b/presentation/rstudio conf 2021/Abstract proposal.md @@ -0,0 +1,5 @@ +Learn how to handle 100GBs of data with ease using {disk.frame} - the larger-than-RAM-data manipulation package. + +R loads data in its entirety into RAM. However, RAM is a precious resource and often do run out. That's why most R user would have run into the "cannot allocate vector of size xxB." error at some point. + +However, the need to handle larger-than-RAM data doesn't go away just because RAM isn't large enough. So many useRs turn to big data tools like Spark for the task. In this talk, I will make the case that {disk.frame} is sufficient and often preferable for manipulating larger-than-RAM data that fit on disk. I will show how you can apply familiar {dplyr}-verbs to manipulate larger-than-RAM data with {disk.frame}. \ No newline at end of file diff --git a/presentation/twin cities/.gitignore b/presentation/twin cities/.gitignore new file mode 100644 index 00000000..26416673 --- /dev/null +++ b/presentation/twin cities/.gitignore @@ -0,0 +1 @@ +*.mp4 diff --git a/tests/testthat/test-add-chunk.r b/tests/testthat/test-add-chunk.r index 07855ac3..c6b38a73 100644 --- a/tests/testthat/test-add-chunk.r +++ b/tests/testthat/test-add-chunk.r @@ -4,6 +4,17 @@ setup({ setup_disk.frame(workers = 2) }) +test_that("guard against github 292", { + a = data.frame(a = as.Date("2020-07-01"), b = runif(1e6)) + + a.df = as.disk.frame(a) + + head(a.df) + + expect_s3_class(add_chunk(a.df, a), "disk.frame") + delete(a.df) +}) + test_that("testing add chunk without naming chunk_id", { a = data.frame(a = 1:100, b = 1:100) @@ -37,5 +48,18 @@ test_that("testing add chunk by naming chunk_id", { delete(a1) }) +test_that("testing add chunk by using compression", { + a = data.frame(a = 1:100, b = 1:100) + + a1 = as.disk.frame(a, overwrite = TRUE) + b = data.frame(a = 51:150, b = 1:100) + d = data.frame(a = 1:50, b = 1:50) + + add_chunk(a1, b, compress=50) + expect_equal(nrow(a1), 200) + + delete(a1) +}) + teardown({ }) \ No newline at end of file diff --git a/tests/testthat/test-names.r b/tests/testthat/test-names.r index 0441136b..2d114aeb 100644 --- a/tests/testthat/test-names.r +++ b/tests/testthat/test-names.r @@ -12,6 +12,13 @@ test_that("testing names", { expect_setequal(names(b), c("a","b")) }) +test_that("testing names with lazyfn", { + b = disk.frame(file.path(tempdir(), "tmp_names.df")) %>% + mutate(d = a + b) + + expect_setequal(colnames(b), c("a","b", "d")) + expect_setequal(names(b), c("a","b", "d")) +}) teardown({ fs::dir_delete(file.path(tempdir(), "tmp_names.df")) diff --git a/vignettes/concepts.Rmd b/vignettes/concepts.Rmd deleted file mode 100644 index d320fd90..00000000 --- a/vignettes/concepts.Rmd +++ /dev/null @@ -1,69 +0,0 @@ ---- -title: "Key `{disk.frame}` concepts" -author: "ZJ" -output: rmarkdown::html_vignette -vignette: > - %\VignetteIndexEntry{Key disk.frame concepts} - %\VignetteEngine{knitr::rmarkdown} - %\VignetteEncoding{UTF-8} ---- - -```{r, include = FALSE} -knitr::opts_chunk$set( - collapse = TRUE, - comment = "#>" -) -``` - -# Key `{disk.frame}` concepts -There are a number of concepts and terminologies that are useful to understand in order to use `disk.frame` effectively. - -## What is a `disk.frame` and what are chunks? - -A `disk.frame` is a folder containing [`fst`](https://www.fstpackage.org/) files named "1.fst", "2.fst", "3.fst" etc. Each of the ".fst" file is called a _chunk_. - -## Workers and parallelism - -Parallelism in `disk.frame` is achieved using the [`future` package](https://cran.r-project.org/package=future). When performing many tasks, `disk.frame` uses multiple workers, where each _worker_ is an R session, to perform the tasks in parallel. - -It is recommended that you should run the following immediately after `library(disk.frame)` to set-up multiple workers. For example: - -```r -library(disk.frame) -setup_disk.frame() - -# this will allow unlimited amount of data to be passed from worker to worker -options(future.globals.maxSize = Inf) -``` - -For example, suppose we wish to compute the number of rows for each chunk, we can clearly perform this simultaneously in parallel. The code to do that is - -```r -# use only one column is fastest -df[,.N, keep = "first_col"] -``` - -or equivalent using the `srckeep` function - -```r -# use only one column is fastest -srckeep(df, "first_col")[,.N, keep = "first_col"] -``` - -Say there are `n` chunks in `df`, and there are `m` workers. Then the first `m` chunks will run `chunk[,.N]` simultaneously. - -To see how many workers are at work, use -```r -# see how many workers are available for work -future::nbrOfWorkers() -``` - -## How `{disk.frame}` works - -When `df %>% some_fn %>% collect` is called. The `some_fn` is applied to each chunk of `df`. The collect will row-bind the results from `some_fn(chunk)`together if the returned value of `some_fn` is a data.frame, or it will return a `list` containing the results of `some_fn`. - -The session that receives these results is called the **main session**. In general, we should try to minimize the amount of data passed from the worker sessions back to the main session, because passing data around can be slow. - -Also, please note that there is no communication between the workers, except for workers passing data back to the main session. - - diff --git a/vignettes/convenience-features.Rmd b/vignettes/convenience-features.Rmd deleted file mode 100644 index 6269a97b..00000000 --- a/vignettes/convenience-features.Rmd +++ /dev/null @@ -1,54 +0,0 @@ ---- -title: "Convenience features" -author: "ZJ" -output: rmarkdown::html_vignette -vignette: > - %\VignetteIndexEntry{Convenience features} - %\VignetteEngine{knitr::rmarkdown} - %\VignetteEncoding{UTF-8} ---- - -```{r, include = FALSE} -knitr::opts_chunk$set( - collapse = TRUE, - comment = "#>" -) -``` - -## Convenience Features - -### GUI for setting options - -I wanted to make `{disk.frame}` as easy to use as possible. I often forget what options are available to me. So I've made a GUI - -```r -setup_disk.frame(gui = TRUE) -``` -which opens up a Shiny app where the user can choose the options. - -### RStudio column name completion - -```r -library(disk.frame) -mtcars.df = as.disk.frame(mtcars) - -mtcars.df %>% - filter() -``` - -you can press tab in RStudio and it will show all column available - -### Insert ceremony/boilerplate into code in RStudio - -The below will insert the recommended ceremony code into your editor -```r -disk.frame::insert_ceremony() -``` -should insert - -```r -# this willl set disk.frame with multiple workers -setup_disk.frame() -# this will allow unlimited amount of data to be passed from worker to worker -options(future.globals.maxSize = Inf) -``` diff --git a/vignettes/data-table-syntax.Rmd b/vignettes/data-table-syntax.Rmd deleted file mode 100644 index e87ca88b..00000000 --- a/vignettes/data-table-syntax.Rmd +++ /dev/null @@ -1,78 +0,0 @@ ---- -title: "Using data.table syntax with disk.frame" -author: "ZJ" -output: rmarkdown::html_vignette -vignette: > - %\VignetteIndexEntry{Using data.table syntax} - %\VignetteEngine{knitr::rmarkdown} - %\VignetteEncoding{UTF-8} ---- - -```{r, include = FALSE} -knitr::opts_chunk$set( - collapse = TRUE, - comment = "#>" -) -``` - -## `disk.frame` supports `data.table` syntax - - -```{r setup_data_table, cache=TRUE} -library(disk.frame) - -# set-up disk.frame to use multiple workers -if(interactive()) { - setup_disk.frame() - # highly recommended, however it is pun into interactive() for CRAN because - # change user options are not allowed on CRAN - options(future.globals.maxSize = Inf) -} else { - setup_disk.frame(2) -} - - -library(nycflights13) - -# create a disk.frame -flights.df = as.disk.frame(nycflights13::flights, outdir = file.path(tempdir(),"flights13"), overwrite = TRUE) -``` - -In the following example, I will use the `.N` from the `data.table` package to count the unique combinations `year` and `month` within each chunk. - -```{r ok, dependson='setup_data_table'} -library(data.table) -library(disk.frame) - -flights.df = disk.frame(file.path(tempdir(),"flights13")) - -names(flights.df) - -flights.df[,.N, .(year, month), keep = c("year", "month")] -``` - -All `data.table` syntax are supported. However, `disk.frame` adds the ability to load only those columns required for the analysis using the `keep =` option. In the above analysis, only the `year` and `month` variables are required and hence `keep = c("year", "month")` was used. - -Alternatively, we can use the `srckeep` function to achieve the same, e.g. - -```r -srckeep(flights.df, c("year", "month"))[,.N, .(year, month)] -``` - -### External variables are captured - -`disk.frame` sends the computation to background workers which are essentially distinct and separate R sessions. Typically, the variables that you have available in your current R session aren't visible in the other R sessions, but `disk.frame` uses the `future` package's variable detection abilities to figure out which variables are in use and then send them to the background workers so they have access to the variables as well. E.g. - -```{r var_detect, dependson='setup_data_table'} -y = 42 -some_fn <- function(x) x - - -flights.df[,some_fn(y)] -``` - -In the above example, neither `some_fn` nor `y` are defined in the background workers' environments, but `disk.frame` still manages to evaluate this code `flights.df[,some_fn(y)]`. - -```{r clean_up, include=FALSE} -fs::dir_delete(file.path(tempdir(),"flights13")) -``` \ No newline at end of file diff --git a/vignettes/glm.Rmd b/vignettes/glm.Rmd deleted file mode 100644 index 1d5fb418..00000000 --- a/vignettes/glm.Rmd +++ /dev/null @@ -1,111 +0,0 @@ ---- -title: "Generalized Linear Models (GLM) including logistic regression with disk.frame" -output: rmarkdown::html_vignette -vignette: > - %\VignetteIndexEntry{Generalized Linear Models (logistic regression etc) with disk.frame} - %\VignetteEngine{knitr::rmarkdown} - %\VignetteEncoding{UTF-8} ---- - -```{r, include = FALSE} -knitr::opts_chunk$set( - collapse = TRUE, - comment = "#>" -) -``` - -```{r setup, cache=TRUE} -suppressPackageStartupMessages(library(disk.frame)) - -if(interactive()) { - setup_disk.frame() -} else { - # only use 1 work to pass CRAN check - setup_disk.frame(1) -} - -``` - -# GLMs - -### Prerequisites -In this article, we will assume you are familiar with Generalized Linear Models (GLMs). You are also expected to have basic working knowledge of {`disk.frame`}, see this [{`disk.frame`} Quick Start](http://diskframe.com/articles/intro-disk-frame.html). - -## Introduction -One can fit a GLM using the `glm` function. For example, - -```{r glm, cache=TRUE} -m = glm(dist ~ speed, data = cars) -``` - -would fit a linear model on the data `cars` with `dist` as the target and `speed` as the explanatory variable. You can inspect the results of the model fit using - -```{r, depeondson='glm'} -summary(m) -``` - -or if you have `{broom}` installed - -```{r, depeondson='glm'} -broom::tidy(m) -``` - -With {`disk.frame`}, you can run GLM `dfglm` function, where the `df` stands for `disk.frame` of course! -```{r dependson='setup'} -cars.df = as.disk.frame(cars) - -m = dfglm(dist ~ speed, cars.df) - -summary(m) - - -majorv = as.integer(version$major) -minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1]) - -if((majorv == 3) & (minorv >= 6)) { - broom::tidy(m) -} else { - # broom doesn't work in version < R3.6 because biglm does not work -} - -``` - -The syntax didn't change at all! You are able to enjoy the benefits of `disk.frame` when dealing with larger-than-RAM data. - -## Logistic regression -Logistic regression is one of the most commonly deployed machine learning (ML) models. It is often used to build binary classification models - -```{r dependson='setup'} -iris.df = as.disk.frame(iris) - -# fit a logistic regression model to predict Speciess == "setosa" using all variables -all_terms_except_species = setdiff(names(iris.df), "Species") -formula_rhs = paste0(all_terms_except_species, collapse = "+") - -formula = as.formula(paste("Species == 'versicolor' ~ ", formula_rhs)) - -iris_model = dfglm(formula , data = iris.df, family=binomial()) - -# iris_model = dfglm(Species == "setosa" ~ , data = iris.df, family=binomial()) - -summary(iris_model) - -majorv = as.integer(version$major) -minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1]) - -if((majorv == 3) & (minorv >= 6)) { - broom::tidy(iris_model) -} else { - # broom doesn't work in version < R3.6 because biglm does not work -} - -``` - -The arguments to the `dfglm` function are the same as the arguments to `biglm::bigglm` which are based on the `glm` function. Please check their documentations for other argument options. - -## Notes -`{disk.frame}` uses `{biglm}` and `{speedglm}` as the backend for GLMs. Unfortunately, neither package is managed on open-source platforms, so it's more difficult to contribute to them by making bug fixes and submitting bug reports. So bugs are likely to persists. There is an active effort on `disk.frame` to look for alternatives. Example of avenues to explore include tighter integration with `{keras}`, h2o, or Julia's OnlineStats.jl for model fit purposes. - -Another package for larger-than-RAM glm fitting, `{bigFastlm}`, has been taken off CRAN, it is managed on Github. - -Currently, parallel processing of GLM fit are not possible with {`disk.frame`}. diff --git a/vignettes/ingesting-data.Rmd b/vignettes/ingesting-data.Rmd deleted file mode 100644 index 541d870a..00000000 --- a/vignettes/ingesting-data.Rmd +++ /dev/null @@ -1,174 +0,0 @@ ---- -title: "Ingesting Data" -author: "ZJ" -output: rmarkdown::html_vignette -vignette: > - %\VignetteIndexEntry{Ingesting data including CSVs} - %\VignetteEngine{knitr::rmarkdown} - %\VignetteEncoding{UTF-8} ---- - -```{r, include = FALSE} -knitr::opts_chunk$set( - collapse = TRUE, - comment = "#>" -) -``` - -# Ingesting Data - -One of the most important tasks to perform before using the `{disk.frame}` package is to make some `disk.frame`s! There are a few functions to help you do that. Before we do that, we set up the `{disk.frame}` as usual - -**Setting up** - -```r -library(disk.frame) - -# set-up disk.frame to use multiple workers -if(interactive()) { - setup_disk.frame() - # highly recommended, however it is pun into interactive() for CRAN because - # change user options are not allowed on CRAN - options(future.globals.maxSize = Inf) -} else { - setup_disk.frame(2) -} - -``` - -## Convert a `data.frame` to `disk.frame` -Firstly, there is `as.disk.frame()` which allows you to make a `disk.frame` from a `data.frame`, e.g. - -```r -flights.df = as.disk.frame(nycflights13::flights) -``` - -will convert the `nycflights13::flights` `data.frame` to a `disk.frame` somewhere in `tempdir()`. To find out the location of the `disk.frame` use: - -```r -attr(flights.df, "path") -``` - -You can also specify a location to output the `disk.frame` to using `outdir` - -```r -flights.df = as.disk.frame(nycflights13::flights, outdir = "some/path.df") -``` - -it is recommended that you use `.df` as the extension for a `disk.frame`, however this is not an enforced requirement. - -However, one of the reasons for `disk.frame` to exist is to handle larger-than-RAM files, hence `as.disk.frame` is not all that useful because it can only convert data that can fit into RAM. `disk.frame` comes with a couple more ways to create `disk.frame`. - -## Creating `disk.frame` from CSVs -The function `csv_to_disk.frame` can convert CSV files to `disk.frame`. The most basic usage is - -```r -some.df = csv_to_disk.frame("some/path.csv", outdir = "some.df") -``` - -this will convert the CSV file `"some/path.csv"` to a `disk.frame`. - -## Multiple CSV files - -However, sometimes we have multiple CSV files that you want to read in and row-bind into one large `disk.frame`. You can do so by supplying a vector of file paths e.g. from the result of `list.files` - -```r -some.df = csv_to_disk.frame(c("some/path/file1.csv", "some/path/file2.csv")) - -# or -some.df = csv_to_disk.frame(list.files("some/path")) -``` - -## Ingesting CSV files chunk-wise -The `csv_to_disk.frame(path, ...)` function reads the file located at `path` in full into RAM but sometimes the CSV file may be too large to read in one go, as that would require loading the whole file into RAM. In that case, you can read the files chunk-by-chunk by using the `in_chunk_size` argument which controls how many rows you read in per chunk - -```r -# to read in 1 million (=1e6) rows per chunk -csv_to_disk.frame(path, in_chunk_size = 1e6) -``` - -When `in_chunk_size` is specified, the input file is split into many smaller files using `bigreadr`'s split file functions. This is generally the fastest way to ingest large CSVs, as the split files can be processed in parallel using all CPU cores. But the disk space requirement is doubled because the split files are as large as the original file. If you run out of disk space, then you must clean R's temporary folder at `tempdir()` and choose another `chunk_reader` e.g. `csv_to_disk.frame(..., chunk_reader = "LaF")`. - -## Sharding -One of the most important aspects of `disk.frame` is sharding. One can shard a `disk.frame` at read time by using the `shardby` - -```r -csv_to_disk.frame(path, shardby = "id") -``` - -In the above case, all rows with the same `id` values will end up in the same chunk. - - -## Just-in-time transformation -Sometimes, one may wish to perform some transformation on the CSV before writing out to disk. One can use the `inmapfn` argument to do that. The `inmapfn` name comes from INput MAPping FuNction. The general usage pattern is as follows: - -```r -csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) { - some_transformation(chunk) -}) -``` - -As a contrived example, suppose you wish to convert a string into date at read time: - -```r -df = data.frame(date_str = c("2019-01-02", "2019-01-02")) - -# write the data.frame -write.csv(df, file.path(tempdir(), "df.csv")) - - -# this would show that date_str is a string -str(collect(csv_to_disk.frame(file.path(tempdir(), "df.csv")))$date_str) -## chr [1:2] "2019-01-02" "2019-01-02" - -# this would show that date_str is a string -df = csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) { - # convert to date_str to date format and store as "date" - chunk[, date := as.Date(date_str, "%Y-%m-%d")] - chunk[, date_str:=NULL] -}) - -str(collect(df)$date) -## Date[1:2], format: "2019-01-02" "2019-01-02" -``` - -## Reading CSVs from zip files -Often, CSV comes zipped in a zip files. You can use the `zip_to_disk.frame` to convert all CSVs within a zip file - -```r -zip_to_disk.frame(path_to_zip_file) -``` - -The arguments for `zip_to_disk.frame` are the same as `csv_to_disk.frame`'s. - - -## Using `add_chunk` - -What if the method of converting to a `disk.frame` isn't implemented in `disk.frame` yet? One can use some lower level constructs provided by `disk.frame` to create `disk.frame`s. For example, the `add_chunk` function can be used to add more chunks to a `disk.frame`, e.g. - -```r -a.df = disk.frame() # create an empty disk.frame -add_chunk(a.df, cars) # adds cars as chunk 1 -add_chunk(a.df, cars) # adds cars as chunk 2 -``` - -Another example of using `add_chunk` is via `readr`'s chunked read functions to create a delimited file reader - -```r -delimited_to_disk.frame <- function(file, outdir, ...) { - res.df = disk.frame(outdir, ...) - readr::read_delim_chunked(file, callback = function(chunk) { - add_chunk(res.df, chunk) - }, ...) - - res.df -} - -delimited_to_disk.frame(path, outdir = "some.df") -``` - -The above code uses `readr`'s `read_delim_chunked` function to read `file` and call `add_chunk`. The problem with this approach is that is it sequential in nature and hence is not able to take advantage of parallelism. - -## Exploiting the structure of a disk.frame - -Of course, a `disk.frame` is just a folder with many `fst` files named as `1.fst`, `2.fst` etc. So one can simply create these `fst` files and ensure they have the same variable names and put them in a folder. \ No newline at end of file diff --git a/vignettes/intro-disk-frame.Rmd b/vignettes/intro-disk-frame.Rmd deleted file mode 100644 index c3d364fb..00000000 --- a/vignettes/intro-disk-frame.Rmd +++ /dev/null @@ -1,378 +0,0 @@ ---- -title: "Quick Start: Basic Operations with nycflights13" -author: "ZJ" -output: rmarkdown::html_vignette -vignette: > - %\VignetteIndexEntry{Quick Start} - %\VignetteEngine{knitr::rmarkdown} - %\VignetteEncoding{UTF-8} ---- - -```{r setup, include = FALSE} -suppressPackageStartupMessages(library(disk.frame)) -library(fst) -library(magrittr) -library(nycflights13) -library(dplyr) -library(data.table) - -# you need to run this for multi-worker support -# limit to 2 cores if not running interactively; most likely on CRAN -# set-up disk.frame to use multiple workers -if(interactive()) { - setup_disk.frame() - # highly recommended, however it is pun into interactive() for CRAN because - # change user options are not allowed on CRAN - options(future.globals.maxSize = Inf) -} else { - setup_disk.frame(2) -} - - -knitr::opts_chunk$set( - eval = FALSE, - collapse = TRUE, - comment = "#>", - include = TRUE -) -``` - -# Quick Start - replicating dplyr's tutorial on nycflight13 - -The [`disk.frame` package](https://github.com/xiaodaigh/disk.frame) aims to be the answer to the question: how do I manipulate structured tabular data that doesn't fit into Random Access Memory (RAM)? - -In a nutshell, `disk.frame` makes use of two simple ideas: - -1) split up a larger-than-RAM dataset into chunks and store each chunk in a separate file inside a folder and -2) provide a convenient API to manipulate these chunks - -`disk.frame` performs a similar role to distributed systems such as Apache Spark, Python's Dask, and Julia's JuliaDB.jl for *medium data* which are datasets that are too large for RAM but not quite large enough to qualify as *big data*. - -In this tutorial, we introduce `disk.frame`, address some common questions, and replicate the [sparklyr data manipulation tutorial](https://spark.rstudio.com/dplyr/) using `disk.frame` constructs. - -## Installation -Simply run - -```r -install.packages("disk.frame") # when CRAN ready -``` -or - -```r -devtools::install_github("xiaodaigh/disk.frame") -``` - -## Set-up `disk.frame` -`disk.frame` works best if it can process multiple data chunks in parallel. The best way to set-up `disk.frame` so that each CPU core runs a background worker is by using - -```r -setup_disk.frame() - -# this will allow unlimited amount of data to be passed from worker to worker -options(future.globals.maxSize = Inf) -``` - -The `setup_disk.frame()` function sets up background workers equal to the number of CPU cores available on your machine; please note that, by default, hyper-threaded cores are counted as one not two. - -Alternatively, one may specify the number of workers using `setup_disk.frame(workers = n)`. - -## Basic Data Operations with `disk.frame` - -The `disk.frame` package provides convenient functions to convert `data.frame`s and CSVs to `disk.frame`s. - -### Creating a `disk.frame` from `data.frame` -We convert a `data.frame` to `disk.frame` using the `as.data.frame` function. - -```{r asdiskframe, cache=TRUE} -library(nycflights13) -library(dplyr) -library(disk.frame) -library(data.table) - -# convert the flights data to a disk.frame and store the disk.frame in the folder -# "tmp_flights" and overwrite any content if needed -flights.df <- as.disk.frame( - flights, - outdir = file.path(tempdir(), "tmp_flights.df"), - overwrite = TRUE) -flights.df -``` -You should now see a folder called `tmp_flights` with some files in it, namely `1.fst`, `2.fst`... where each `fst` files is one chunk of the `disk.frame`. - - -### Creating a `disk.frame` from CSV -```{r} -library(nycflights13) -# write a csv -csv_path = file.path(tempdir(), "tmp_flights.csv") -data.table::fwrite(flights, csv_path) - -# load the csv into a disk.frame -df_path = file.path(tempdir(), "tmp_flights.df") -flights.df <- csv_to_disk.frame( - csv_path, - outdir = df_path, - overwrite = T) - -flights.df -``` - -If the CSV is too large to read in, then we can also use the `in_chunk_size` option to control how many rows to read in at once. For example, to read in the data 100,000 rows at a time: - -```{r} -library(nycflights13) -library(disk.frame) - -# write a csv -csv_path = file.path(tempdir(), "tmp_flights.csv") - -data.table::fwrite(flights, csv_path) - -df_path = file.path(tempdir(), "tmp_flights.df") - -flights.df <- csv_to_disk.frame( - csv_path, - outdir = df_path, - in_chunk_size = 100000) - -flights.df -``` - -`disk.frame` also has a function `zip_to_disk.frame` that can convert every CSV in a zip file to a `disk.frame`. - -### Simple `dplyr` verbs and lazy evaluation -```{r dfselect, dependson='asdiskframe', cache=TRUE} -flights.df1 <- select(flights.df, year:day, arr_delay, dep_delay) -flights.df1 -``` - -```{r dependson='dfselect'} -class(flights.df1) -``` - -The class of `flights.df1` is also a `disk.frame` after the `dplyr::select` transformation. Also, `disk.frame` operations are by default (and where possible) **lazy**, meaning they don't perform the operations right away. Instead, these functions wait until you call `collect`. Exceptions to this rule are the `*_join` operations, which evaluate *eagerly* under certain conditions--see **Joins for disk.frame in-depth** for details. - -For lazily constructed `disk.frame`s (e.g. `flights.df1`), the function `collect` can be used to bring the results from disk into R, e.g. -```{r, dependson='dfselect'} -collect(flights.df1) %>% head(2) -``` - -Of course, for larger-than-RAM datasets, one wouldn't call `collect` on the whole `disk.frame` (because why would you need `disk.frame` otherwise). More likely, one would call `collect` on a `filter`ed dataset or one summarized with `group_by`. - -Some examples of other dplyr verbs applied: - -```{r, dependson='asdiskframe'} -filter(flights.df, dep_delay > 1000) %>% collect %>% head(2) -``` - -```{r, dependson='asdiskframe'} -mutate(flights.df, speed = distance / air_time * 60) %>% collect %>% head(2) -``` - -### Examples of NOT fully supported `dplyr` verbs - -The `chunk_arrange` function arranges (sorts) each chunk but not the whole dataset. So use with caution. Similarly, `chunk_summarise` creates summary variables within each chunk and hence also needs to be used with caution. In the **Group-by** section, we demonstrate how to use `summarise` in the `disk.frame` context correctly with `hard_group_by`s. - -```{r, dependson='asdiskframe'} -# this only sorts within each chunk -chunk_arrange(flights.df, dplyr::desc(dep_delay)) %>% collect %>% head(2) -``` - - -```{r, dependson='asdiskframe'} -chunk_summarize(flights.df, mean_dep_delay = mean(dep_delay, na.rm =T)) %>% collect -``` - -### Piping - -One can chain `dplyr` verbs together like with a `data.frame` - -```{r, dependson='asdiskframe'} -c4 <- flights %>% - filter(month == 5, day == 17, carrier %in% c('UA', 'WN', 'AA', 'DL')) %>% - select(carrier, dep_delay, air_time, distance) %>% - mutate(air_time_hours = air_time / 60) %>% - collect %>% - arrange(carrier)# arrange should occur after `collect` - -c4 %>% head -``` - -### List of supported `dplyr` verbs - -```r -select -rename -filter -chunk_arrange # within each chunk -chunk_group_by # within each chunk -chunk_summarize # within each chunk -group_by # limited functions -summarize # limited functions -mutate -transmute -left_join -inner_join -full_join # careful. Performance! -semi_join -anit_join -``` - -## Sharding and distribution of chunks - -Like other distributed data manipulation frameworks, `disk.frame` utilizes the *sharding* concept to distribute the data into chunks. For example, "to shard by `cust_id`" means that all rows with the same `cust_id` will be stored in the same chunk. This enables `chunk_group_by` by `cust_id` to produce the same results as non-chunked data. - -The `by` variables that were used to shard the dataset are called the `shardkey`s. The *sharding* is performed by computing a deterministic hash on the shard keys (the `by` variables) for each row. The hash function produces an integer between `1` and `n`, where `n` is the number of chunks. - -## Group-by - -`{disk.frame}` implements the `group_by` operation with some caveats. In the `{disk.frame}` framework, only a subset of functions are supported in `summarize`. However, the user can create more custom `group-by` functions on the fly. - -```{r, dependson='asdiskframe'} -flights.df %>% - group_by(carrier) %>% # notice that hard_group_by needs to be set - summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>% # mean follows normal R rules - collect %>% - arrange(carrier) -``` - -## Restrict input columns for faster processing - -One can restrict which input columns to load into memory for each chunk; this can significantly increase the speed of data processing. To restrict the input columns, use the `srckeep` function which only accepts column names as a string vector. - -```{r, dependson='asdiskframe'} -flights.df %>% - srckeep(c("carrier","dep_delay")) %>% - group_by(carrier) %>% - summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>% # mean follows normal R rules - collect -``` - -Input column restriction is one of the most critical efficiencies provided by `disk.frame`. Because the underlying format allows random access to columns (i.e. retrieve only the columns used for processing), hence one can drastically reduce the amount of data loaded into RAM for processing by keeping only those columns that are directly used to produce the results. - -## Joins - -`disk.frame` supports many dplyr joins including: - -```r -left_join -inner_join -semi_join -inner_join -full_join # requires hard_group_by on both left and right -``` -In all cases, the left dataset (`x`) must be a `disk.frame`, and the right dataset (`y`) can be either a `disk.frame` or a `data.frame`. If the right dataset is a `disk.frame` and the `shardkey`s are different between the two `disk.frame`s then two expensive `hard` `group_by` operations are performed *eagerly*, one on the left `disk.frame` and one on the right `disk.frame` to perform the joins correctly. - -However, if the right dataset is a `data.frame` then `hard_group_by`s are only performed in the case of `full_join`. - -Note `disk.frame` does not support `right_join`. The user should use `left_join` instead. - -The below joins are performed *lazily* because `airlines.dt` is a `data.table` not a `disk.frame`: - -```{r airlines_dt, dependson='asdiskframe', cache=TRUE} -# make airlines a data.table -airlines.dt <- data.table(airlines) -# flights %>% left_join(airlines, by = "carrier") # -flights.df %>% - left_join(airlines.dt, by ="carrier") %>% - collect %>% - head -``` - -```{r, dependson='airlines_dt'} -flights.df %>% - left_join(airlines.dt, by = c("carrier", "carrier")) %>% - collect %>% - tail -``` - -## Window functions and arbitrary functions - -`{disk.frame}` supports all `data.frame` operations, unlike Spark which can only perform those operations that Spark has implemented. Hence windowing functions like `min_rank` and `rank` are supported out of the box. - -For the following example, we will use the `hard_group_by` which performs a group-by and also reorganises the chunks so that all records with the same `year`, `month`, and `day` end up in the same chunk. This is typically not advised, as `hard_group_by` can be slow for large datasets. - -```{r, dependson='asdiskframe'} -# Find the most and least delayed flight each day -bestworst <- flights.df %>% - srckeep(c("year","month","day", "dep_delay")) %>% - hard_group_by(c("year", "month", "day")) %>% - filter(dep_delay == min(dep_delay, na.rm = T) || dep_delay == max(dep_delay, na.rm = T)) %>% - collect - -bestworst %>% head -``` - -Another example: - -```{r, dependson='asdiskframe'} -ranked <- flights.df %>% - srckeep(c("year","month","day", "dep_delay")) %>% - hard_group_by(c("year", "month", "day")) %>% - filter(min_rank(desc(dep_delay)) <= 2 & dep_delay > 0) %>% - collect - -ranked %>% head -``` - -One more example: - -```{r, dependson='asdiskframe'} -# Rank each flight within a daily window -ranked <- flights.df %>% - srckeep(c("year","month","day", "dep_delay")) %>% - chunk_group_by(year, month, day) %>% - select(dep_delay) %>% - mutate(rank = rank(desc(dep_delay))) %>% - collect - -ranked %>% head -``` - - -## Arbitrary by-chunk processing - -One can apply arbitrary transformations to each chunk of the `disk.frame` by using the `delayed` function which evaluates lazily or the `map.disk.frame(lazy = F)` function which evaluates eagerly. For example to return the number of rows in each chunk: - -```{r, dependson='asdiskframe'} -flights.df1 <- delayed(flights.df, ~nrow(.x)) -collect_list(flights.df1) %>% head # returns number of rows for each data.frame in a list -``` -and to do the same with `map.disk.frame`: - -```{r, dependson='asdiskframe'} -map(flights.df, ~nrow(.x), lazy = F) %>% head -``` -The `map` function can also output the results to another disk.frame folder, e.g. - -```{r, dependson='asdiskframe'} -# return the first 10 rows of each chunk -flights.df2 <- map(flights.df, ~.x[1:10,], lazy = F, outdir = file.path(tempdir(), "tmp2"), overwrite = T) - -flights.df2 %>% head -``` - -Notice `{disk.frame}` supports the `purrr` syntax for defining a function using `~`. - -## Sampling - -In the `disk.frame` framework, sampling a proportion of rows within each chunk can be performed using `sample_frac`. - -```{r, dependson='asdiskframe'} -flights.df %>% sample_frac(0.01) %>% collect %>% head -``` - -## Writing Data - -One can output a `disk.frame` by using the `write_disk.frame` function. E.g. - -```r -write_disk.frame(flights.df, outdir="out") -``` -this will output a disk.frame to the folder "out" - -```{r cleanup} -fs::dir_delete(file.path(tempdir(), "tmp_flights.df")) -fs::dir_delete(file.path(tempdir(), "tmp2")) -fs::file_delete(file.path(tempdir(), "tmp_flights.csv")) -``` From 511f76b2695decb3b902752f393da197474f116f Mon Sep 17 00:00:00 2001 From: evalparse Date: Thu, 13 Aug 2020 14:13:12 +1000 Subject: [PATCH 3/8] Update README.Rmd --- README.Rmd | 22 ++++++++++++---------- 1 file changed, 12 insertions(+), 10 deletions(-) diff --git a/README.Rmd b/README.Rmd index af17f732..480291fb 100644 --- a/README.Rmd +++ b/README.Rmd @@ -270,17 +270,19 @@ The work priorities at this stage are ## Blogs and other resources -| Title | Language | Author | Date | Description | -| ------------------------------------------------------------ | -------- | --------------- | ---------- | ------------------------------------------------------------ | -| [25 days of disk.frame](https://twitter.com/evalparse/status/1200963268270886912) | English | ZJ | 2019-12-01 | 25 tweets about `{disk.frame}` | -| https://www.researchgate.net/post/What_is_the_Maximum_size_of_data_that_is_supported_by_R-datamining | English | Knut Jägersberg | 2019-11-11 | Great answer on using disk.frame | -| [`{disk.frame}` is epic](https://www.brodrigues.co/blog/2019-09-03-disk_frame/) | English | Bruno Rodriguez | 2019-09-03 | It's about loading a 30G file into `{disk.frame}` | -| [My top 10 R packages for data analytics](https://www.actuaries.digital/2019/09/26/my-top-10-r-packages-for-data-analytics/) | English | Jacky Poon | 2019-09-03 | `{disk.frame}` was number 3 | -| [useR! 2019 presentation video](https://www.youtube.com/watch?v=3XMTyi_H4q4) | English | Dai ZJ | 2019-08-03 | | -| [useR! 2019 presentation slides](https://www.beautiful.ai/player/-LphQ0YaJwRektb8nZoY) | English | Dai ZJ | 2019-08-03 | | +| Title | Language | Author | Date | Description | +|---------------------------------------------------------------------------------------------------------------------------------------|----------|-----------------|------------|----------------------------------------------------------------------------------------------------| +| [useR! 2020 Rladies San Diego {disk.frame} tutorial](https://www.youtube.com/watch?v=kjPjXs0mkwE) | English | Dai ZJ | 2020 | A 2 hour {disk.frame} tutorial for beginners | +| [25 days of disk.frame](https://twitter.com/evalparse/status/1200963268270886912) | English | ZJ | 2019-12-01 | 25 tweets about `{disk.frame}` | +| https://www.researchgate.net/post/What_is_the_Maximum_size_of_data_that_is_supported_by_R-datamining | English | Knut Jägersberg | 2019-11-11 | Great answer on using disk.frame | +| [`{disk.frame}` is epic](https://www.brodrigues.co/blog/2019-09-03-disk_frame/) | English | Bruno Rodriguez | 2019-09-03 | It's about loading a 30G file into `{disk.frame}` | +| [My top 10 R packages for data analytics](https://www.actuaries.digital/2019/09/26/my-top-10-r-packages-for-data-analytics/) | English | Jacky Poon | 2019-09-03 | `{disk.frame}` was number 3 | +| [useR! 2019 presentation video](https://www.youtube.com/watch?v=3XMTyi_H4q4) | English | Dai ZJ | 2019-08-03 | | +| [useR! 2019 presentation slides](https://www.beautiful.ai/player/-LphQ0YaJwRektb8nZoY) | English | Dai ZJ | 2019-08-03 | | | [Split-apply-combine for Maximum Likelihood Estimation of a linear model](https://www.brodrigues.co/blog/2019-10-05-parallel_maxlik/) | English | Bruno Rodriguez | 2019-10-06 | `{disk.frame}` used in helping to create a maximum likelihood estimation program for linear models | -| [Emma goes to useR! 2019](https://emmavestesson.netlify.com/2019/07/user2019/) | English | Emma Vestesson | 2019-07-16 | The first mention of `{disk.frame}` in a blog post | -| [深入对比数据科学工具箱：Python3 和 R 之争(2020版)](https://segmentfault.com/a/1190000021653567) | Chinese | Harry Zhu | 2020-02-16 | Mentions disk.frame | +| [Emma goes to useR! 2019](https://emmavestesson.netlify.com/2019/07/user2019/) | English | Emma Vestesson | 2019-07-16 | The first mention of `{disk.frame}` in a blog post | +| [深入对比数据科学工具箱：Python3 和 R 之争(2020版)](https://segmentfault.com/a/1190000021653567) | Chinese | Harry Zhu | 2020-02-16 | Mentions disk.frame | + ### Interested in learning `{disk.frame}` in a structured course? From a3da2c1ffe2dc9d547cce9a71f289807de3eab14 Mon Sep 17 00:00:00 2001 From: xiaodaigh Date: Wed, 7 Oct 2020 01:58:47 +1100 Subject: [PATCH 4/8] trouble shooting guide --- book/88-trouble-shooting.Rmd | 53 ++++++++++++++++++++++++++++++++++++ 1 file changed, 53 insertions(+) create mode 100644 book/88-trouble-shooting.Rmd diff --git a/book/88-trouble-shooting.Rmd b/book/88-trouble-shooting.Rmd new file mode 100644 index 00000000..f9cbca51 --- /dev/null +++ b/book/88-trouble-shooting.Rmd @@ -0,0 +1,53 @@ +--- +title: "Trouble shooting" +author: "ZJ" +output: pdf_document +--- + +```{r include=FALSE} +knitr::opts_chunk$set( + collapse = TRUE, + comment = "#>", + eval=TRUE, + include=TRUE +) +``` + +### Steps to trouble shoot + +1. I suggest updating {future} and your R version if you have not already done so. + +2. Are you able to share the data? + +3. Do a good MWE +``` +library(disk.frame) +setup_disk.frame() + +df<-as.disk.frame(a) + + +df1 = mutate(df, date = as.Date(as.character(datadate), format="%Y%m%d")) + +head(df1) +``` + + +3. Check if your virus scanner is blocking interprocess communication + +4. Try to apply the function to just one chunk, perhaps there is a syntax error or column error? If one chunk works then you can rule out coding error + +``` +get_chunk(df, 1) %>% + mutate(date = as.Date(as.character(datadate), format="%Y%m%d")) +``` + +5. Set the number of workers to 1, so there is no more inter-process communication. Does it work now? If it does, then it's the inter process communication. You might need to contact your admin for help + +``` +setup_disk.frame(workers=1) +mutate(df, date = as.Date(as.character(datadate), format="%Y%m%d")) +As an MWE this works for me. + +a = data.frame(datadate = rep("20201007", 3e6)) +``` From 34bafaab2d60d3251fe43d4997fa27aecbab975b Mon Sep 17 00:00:00 2001 From: xiaodaigh Date: Thu, 18 Feb 2021 10:56:31 +1100 Subject: [PATCH 5/8] updated read me --- README.Rmd | 21 +++++++++---------- README.md | 61 +++++++++++++++++++++++++++--------------------------- 2 files changed, 40 insertions(+), 42 deletions(-) diff --git a/README.Rmd b/README.Rmd index 480291fb..b410c36e 100644 --- a/README.Rmd +++ b/README.Rmd @@ -270,18 +270,17 @@ The work priorities at this stage are ## Blogs and other resources -| Title | Language | Author | Date | Description | -|---------------------------------------------------------------------------------------------------------------------------------------|----------|-----------------|------------|----------------------------------------------------------------------------------------------------| -| [useR! 2020 Rladies San Diego {disk.frame} tutorial](https://www.youtube.com/watch?v=kjPjXs0mkwE) | English | Dai ZJ | 2020 | A 2 hour {disk.frame} tutorial for beginners | -| [25 days of disk.frame](https://twitter.com/evalparse/status/1200963268270886912) | English | ZJ | 2019-12-01 | 25 tweets about `{disk.frame}` | -| https://www.researchgate.net/post/What_is_the_Maximum_size_of_data_that_is_supported_by_R-datamining | English | Knut Jägersberg | 2019-11-11 | Great answer on using disk.frame | -| [`{disk.frame}` is epic](https://www.brodrigues.co/blog/2019-09-03-disk_frame/) | English | Bruno Rodriguez | 2019-09-03 | It's about loading a 30G file into `{disk.frame}` | -| [My top 10 R packages for data analytics](https://www.actuaries.digital/2019/09/26/my-top-10-r-packages-for-data-analytics/) | English | Jacky Poon | 2019-09-03 | `{disk.frame}` was number 3 | -| [useR! 2019 presentation video](https://www.youtube.com/watch?v=3XMTyi_H4q4) | English | Dai ZJ | 2019-08-03 | | -| [useR! 2019 presentation slides](https://www.beautiful.ai/player/-LphQ0YaJwRektb8nZoY) | English | Dai ZJ | 2019-08-03 | | +| Title | Language | Author | Date | Description | +| ------------------------------------------------------------ | -------- | --------------- | ---------- | ------------------------------------------------------------ | +| [25 days of disk.frame](https://twitter.com/evalparse/status/1200963268270886912) | English | ZJ | 2019-12-01 | 25 tweets about `{disk.frame}` | +| https://www.researchgate.net/post/What-is-the-Maximum-size-of-data-that-is-supported-by-R-datamining | English | Knut Jägersberg | 2019-11-11 | Great answer on using disk.frame | +| [`{disk.frame}` is epic](https://www.brodrigues.co/blog/2019-09-03-disk_frame/) | English | Bruno Rodriguez | 2019-09-03 | It's about loading a 30G file into `{disk.frame}` | +| [My top 10 R packages for data analytics](https://www.actuaries.digital/2019/09/26/my-top-10-r-packages-for-data-analytics/) | English | Jacky Poon | 2019-09-03 | `{disk.frame}` was number 3 | +| [useR! 2019 presentation video](https://www.youtube.com/watch?v=3XMTyi_H4q4) | English | Dai ZJ | 2019-08-03 | | +| [useR! 2019 presentation slides](https://www.beautiful.ai/player/-LphQ0YaJwRektb8nZoY) | English | Dai ZJ | 2019-08-03 | | | [Split-apply-combine for Maximum Likelihood Estimation of a linear model](https://www.brodrigues.co/blog/2019-10-05-parallel_maxlik/) | English | Bruno Rodriguez | 2019-10-06 | `{disk.frame}` used in helping to create a maximum likelihood estimation program for linear models | -| [Emma goes to useR! 2019](https://emmavestesson.netlify.com/2019/07/user2019/) | English | Emma Vestesson | 2019-07-16 | The first mention of `{disk.frame}` in a blog post | -| [深入对比数据科学工具箱：Python3 和 R 之争(2020版)](https://segmentfault.com/a/1190000021653567) | Chinese | Harry Zhu | 2020-02-16 | Mentions disk.frame | +| [Emma goes to useR! 2019](https://emmavestesson.netlify.app/2019/07/user2019/) | English | Emma Vestesson | 2019-07-16 | The first mention of `{disk.frame}` in a blog post | +| [深入对比数据科学工具箱：Python3 和 R 之争(2020版)](https://segmentfault.com/a/1190000021653567) | Chinese | Harry Zhu | 2020-02-16 | Mentions disk.frame | diff --git a/README.md b/README.md index 343f76bf..d7e33912 100644 --- a/README.md +++ b/README.md @@ -8,13 +8,13 @@ How do I manipulate tabular data that doesn’t fit into Random Access Memory (RAM)? -Use `{disk.frame}`\! +Use `{disk.frame}`! In a nutshell, `{disk.frame}` makes use of two simple ideas -1) split up a larger-than-RAM dataset into chunks and store each chunk +1. split up a larger-than-RAM dataset into chunks and store each chunk in a separate file inside a folder and -2) provide a convenient API to manipulate these chunks +2. provide a convenient API to manipulate these chunks `{disk.frame}` performs a similar role to distributed systems such as Apache Spark, Python’s Dask, and Julia’s JuliaDB.jl for *medium data* @@ -48,28 +48,28 @@ install.packages("disk.frame", repo="https://cran.rstudio.com") Please see these vignettes and articles about `{disk.frame}` - - [Quick start: +- [Quick start: `{disk.frame}`](https://diskframe.com/articles/intro-disk-frame.html) which replicates the `sparklyr` vignette for manipulating the `nycflights13` flights data. - - [Ingesting data into +- [Ingesting data into `{disk.frame}`](https://diskframe.com/articles/ingesting-data.html) which lists some commons way of creating disk.frames - - [`{disk.frame}` can be more - epic\!](https://diskframe.com/articles/more-epic.html) shows some +- [`{disk.frame}` can be more + epic!](https://diskframe.com/articles/more-epic.html) shows some ways of loading large CSVs and the importance of `srckeep` - - [Group-by](https://diskframe.com/articles/group-by.html) the various +- [Group-by](https://diskframe.com/articles/group-by.html) the various types of group-bys - - [Custom one-stage group-by +- [Custom one-stage group-by functions](https://diskframe.com/articles/custom-group-by.html) how to define custom one-stage group-by functions - - [Fitting GLMs (including logistic +- [Fitting GLMs (including logistic regression)](https://diskframe.com/articles/glm.html) introduces the `dfglm` function for fitting generalized linear models - - [Using data.table syntax with +- [Using data.table syntax with disk.frame](https://diskframe.com/articles/data-table-syntax.html) - - [disk.frame concepts](https://diskframe.com/articles/concepts.html) - - [Benchmark 1: disk.frame vs Dask vs +- [disk.frame concepts](https://diskframe.com/articles/concepts.html) +- [Benchmark 1: disk.frame vs Dask vs JuliaDB](https://diskframe.com/articles/vs-dask-juliadb.html) ## Common questions @@ -256,7 +256,7 @@ limitation that function that depend on the order a column can only be obtained using estimated methods. | Function | Exact/Estimate | Notes | -| ------------ | -------------- | ------------------------------------------ | +|--------------|----------------|--------------------------------------------| | `min` | Exact | | | `max` | Exact | | | `mean` | Exact | | @@ -324,7 +324,7 @@ To find out where the disk.frame is stored on disk: ``` r # where is the disk.frame stored attr(flights.df, "path") -#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpsnJlFJ\\file3d3ce978e3.df" +#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpMpwOj2\\file4bb025d75c63.df" ``` A number of data.frame functions are implemented for disk.frame @@ -387,16 +387,16 @@ The work priorities at this stage are ## Blogs and other resources | Title | Language | Author | Date | Description | -| ------------------------------------------------------------------------------------------------------------------------------------- | -------- | --------------- | ---------- | -------------------------------------------------------------------------------------------------- | +|---------------------------------------------------------------------------------------------------------------------------------------|----------|-----------------|------------|----------------------------------------------------------------------------------------------------| | [25 days of disk.frame](https://twitter.com/evalparse/status/1200963268270886912) | English | ZJ | 2019-12-01 | 25 tweets about `{disk.frame}` | -| | English | Knut Jägersberg | 2019-11-11 | Great answer on using disk.frame | +| | English | Knut Jägersberg | 2019-11-11 | Great answer on using disk.frame | | [`{disk.frame}` is epic](https://www.brodrigues.co/blog/2019-09-03-disk_frame/) | English | Bruno Rodriguez | 2019-09-03 | It’s about loading a 30G file into `{disk.frame}` | | [My top 10 R packages for data analytics](https://www.actuaries.digital/2019/09/26/my-top-10-r-packages-for-data-analytics/) | English | Jacky Poon | 2019-09-03 | `{disk.frame}` was number 3 | -| [useR\! 2019 presentation video](https://www.youtube.com/watch?v=3XMTyi_H4q4) | English | Dai ZJ | 2019-08-03 | | -| [useR\! 2019 presentation slides](https://www.beautiful.ai/player/-LphQ0YaJwRektb8nZoY) | English | Dai ZJ | 2019-08-03 | | +| [useR! 2019 presentation video](https://www.youtube.com/watch?v=3XMTyi_H4q4) | English | Dai ZJ | 2019-08-03 | | +| [useR! 2019 presentation slides](https://www.beautiful.ai/player/-LphQ0YaJwRektb8nZoY) | English | Dai ZJ | 2019-08-03 | | | [Split-apply-combine for Maximum Likelihood Estimation of a linear model](https://www.brodrigues.co/blog/2019-10-05-parallel_maxlik/) | English | Bruno Rodriguez | 2019-10-06 | `{disk.frame}` used in helping to create a maximum likelihood estimation program for linear models | -| [Emma goes to useR\! 2019](https://emmavestesson.netlify.com/2019/07/user2019/) | English | Emma Vestesson | 2019-07-16 | The first mention of `{disk.frame}` in a blog post | -| [深入对比数据科学工具箱：Python3 和 R 之争(2020版)](https://segmentfault.com/a/1190000021653567) | Chinese | Harry Zhu | 2020-02-16 | Mentions disk.frame | +| [Emma goes to useR! 2019](https://emmavestesson.netlify.app/2019/07/user2019/) | English | Emma Vestesson | 2019-07-16 | The first mention of `{disk.frame}` in a blog post | +| [深入对比数据科学工具箱：Python3 和 R 之争(2020版)](https://segmentfault.com/a/1190000021653567) | Chinese | Harry Zhu | 2020-02-16 | Mentions disk.frame | ### Interested in learning `{disk.frame}` in a structured course? @@ -412,7 +412,7 @@ perhaps you have a feature request? Please consider sponsoring ### Backers -Thank you to all our backers\! +Thank you to all our backers! @@ -427,7 +427,7 @@ show up here with a link to your website. **Do you need help with machine learning and data science in R, Python, or Julia?** I am available for Machine Learning/Data -Science/R/Python/Julia consulting\! [Email +Science/R/Python/Julia consulting! [Email me](mailto:dzj@analytixware.com) ## Non-financial ways to contribute @@ -435,16 +435,16 @@ me](mailto:dzj@analytixware.com) Do you wish to give back the open-source community in non-financial ways? Here are some ways you can contribute - - Write a blogpost about your `{disk.frame}`. I would love to learn - more about how `{disk.frame}` has helped you - - Tweet or post on social media (e.g LinkedIn) about `{disk.frame}` to +- Write a blogpost about your `{disk.frame}` usage or experience. I + would love to learn more about how `{disk.frame}` has helped you +- Tweet or post on social media (e.g LinkedIn) about `{disk.frame}` to help promote it - - Bring attention to typos and grammatical errors by correcting and +- Bring attention to typos and grammatical errors by correcting and making a PR. Or simply by [raising an issue here](https://github.com/xiaodaigh/disk.frame/issues) - - Star the [`{disk.frame}` Github +- Star the [`{disk.frame}` Github repo](https://github.com/xiaodaigh/disk.frame) - - Star any repo that `{disk.frame}` depends on +- Star any repo that `{disk.frame}` depends on e.g. [`{fst}`](https://github.com/fstpackage/fst) and [`{future}`](https://github.com/HenrikBengtsson/future) @@ -452,7 +452,6 @@ ways? Here are some ways you can contribute - @@ -467,4 +466,4 @@ status](https://ci.appveyor.com/api/projects/status/github/xiaodaigh/disk.frame? ## Live Stream of `{disk.frame}` development - - +- From e9af01639a28e4b1a5288a0b03a54c73eb94f8c8 Mon Sep 17 00:00:00 2001 From: xiaodaigh Date: Sun, 9 May 2021 20:07:31 +1000 Subject: [PATCH 6/8] updated --- CRAN-RELEASE | 4 +- DESCRIPTION | 4 +- NAMESPACE | 2 - NEWS.md | 6 + R/dplyr_verbs.r | 8 +- R/recommend_nchunks.r | 16 +- README.md | 45 +- book/01-intro.Rmd | 2 +- book/06-vs-dask-juliadb.Rmd | 2 +- cran-comments.md | 12 +- docs/404.html | 11 +- docs/LICENSE-text.html | 11 +- docs/articles/01-intro.html | 13 +- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/02-intro-disk-frame.html | 502 +++++++++--------- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/03-concepts.html | 39 +- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/04-ingesting-data.html | 110 ++-- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/05-data-table-syntax.html | 57 +- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/06-vs-dask-juliadb.html | 166 +++--- .../figure-html/unnamed-chunk-2-1.png | Bin 15944 -> 28103 bytes .../figure-html/unnamed-chunk-3-1.png | Bin 16175 -> 27085 bytes .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/07-glm.html | 93 ++-- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/08-more-epic.html | 203 +++---- .../figure-html/unnamed-chunk-2-1.png | Bin 19434 -> 38731 bytes .../figure-html/unnamed-chunk-3-1.png | Bin 14286 -> 32146 bytes .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/09-convenience-features.html | 40 +- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/10-group-by.html | 149 +++--- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/11-custom-group-by.html | 155 +++--- .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/88-trouble-shooting.html | 186 +++++++ .../header-attrs-2.8/header-attrs.js | 12 + docs/articles/index.html | 13 +- docs/authors.html | 11 +- docs/index.html | 247 +++++---- docs/news/index.html | 47 +- docs/pkgdown.css | 4 +- docs/pkgdown.yml | 7 +- docs/reference/add_chunk.html | 49 +- docs/reference/as.data.frame.disk.frame.html | 21 +- docs/reference/as.data.table.disk.frame.html | 24 +- docs/reference/as.disk.frame.html | 42 +- docs/reference/chunk_group_by.html | 19 +- docs/reference/cmap.html | 162 +++--- docs/reference/cmap2.html | 29 +- docs/reference/collect.html | 39 +- docs/reference/colnames.html | 33 +- docs/reference/compute.disk.frame.html | 36 +- docs/reference/create_chunk_mapper.html | 91 ++-- docs/reference/csv_to_disk.frame.html | 54 +- docs/reference/delete.html | 18 +- docs/reference/df_ram_size.html | 16 +- docs/reference/dfglm.html | 38 +- docs/reference/disk.frame.html | 29 +- docs/reference/dplyr_verbs.html | 62 ++- docs/reference/evalparseglue.html | 13 +- docs/reference/foverlaps.disk.frame.html | 57 +- docs/reference/gen_datatable_synthetic.html | 13 +- docs/reference/get_chunk.html | 32 +- docs/reference/get_chunk_ids.html | 37 +- docs/reference/group_by.html | 21 +- docs/reference/groups.disk.frame.html | 13 +- docs/reference/hard_arrange.html | 45 +- docs/reference/hard_group_by.html | 53 +- docs/reference/head_tail.html | 26 +- docs/reference/index.html | 19 +- docs/reference/is_disk.frame.html | 22 +- docs/reference/join.html | 175 +++--- docs/reference/make_glm_streaming_fn.html | 40 +- docs/reference/merge.disk.frame.html | 42 +- docs/reference/move_to.html | 26 +- docs/reference/nchunks.html | 28 +- docs/reference/ncol_nrow.html | 28 +- docs/reference/one-stage-group-by-verbs.html | 67 +-- docs/reference/overwrite_check.html | 22 +- docs/reference/print.disk.frame.html | 13 +- docs/reference/pull.disk.frame.html | 13 +- docs/reference/rbindlist.disk.frame.html | 39 +- docs/reference/rechunk.html | 46 +- docs/reference/recommend_nchunks.html | 31 +- docs/reference/remove_chunk.html | 36 +- docs/reference/sample.html | 72 +-- docs/reference/setup_disk.frame.html | 39 +- docs/reference/shard.html | 41 +- docs/reference/shardkey.html | 13 +- docs/reference/shardkey_equal.html | 13 +- docs/reference/show_ceremony.html | 19 +- docs/reference/srckeep.html | 23 +- docs/reference/sub-.disk.frame.html | 21 +- docs/reference/tbl_vars.disk.frame.html | 15 +- docs/reference/write_disk.frame.html | 53 +- docs/reference/zip_to_disk.frame.html | 45 +- man/dplyr_verbs.Rd | 3 - tests/testthat.R | 4 - tests/testthat/test-Rcpp.R | 6 - tests/testthat/test-add-chunk.r | 65 --- tests/testthat/test-anti_join.R | 75 --- tests/testthat/test-as-data-frame.R | 14 - tests/testthat/test-as-disk-frame.R | 15 - tests/testthat/test-bloom-filter.r | 16 - tests/testthat/test-collect.R | 57 -- tests/testthat/test-compute.r | 55 -- tests/testthat/test-csv2disk.frame.r | 84 --- tests/testthat/test-data-table.r | 41 -- tests/testthat/test-delete.r | 20 - tests/testthat/test-disk-frame.r | 34 -- tests/testthat/test-dplyr-verbs.r | 175 ------ tests/testthat/test-dtplyr-support.r | 46 -- tests/testthat/test-foverlaps.r | 42 -- tests/testthat/test-full_join.R | 58 -- tests/testthat/test-get_chunk.r | 17 - tests/testthat/test-get_chunk_ids.r | 20 - tests/testthat/test-glm.r | 30 -- tests/testthat/test-group-by.R | 309 ----------- tests/testthat/test-hard-arrange.R | 111 ---- tests/testthat/test-inner_join.R | 64 --- tests/testthat/test-is-disk.frame.r | 12 - tests/testthat/test-left_join.R | 71 --- tests/testthat/test-map.r | 76 --- tests/testthat/test-map2.r | 46 -- tests/testthat/test-merge.r | 59 -- tests/testthat/test-names.r | 25 - tests/testthat/test-nchunks.r | 18 - tests/testthat/test-nrow-ncol.R | 20 - tests/testthat/test-overwrite_check.r | 21 - tests/testthat/test-pls-add.r | 44 -- tests/testthat/test-print.r | 14 - tests/testthat/test-pull.r | 50 -- tests/testthat/test-rbindlist.r | 31 -- tests/testthat/test-rechunk.r | 82 --- tests/testthat/test-recommend_nchunk.R | 16 - tests/testthat/test-remove_chunk.r | 18 - tests/testthat/test-right_join.r | 7 - tests/testthat/test-sample_frac.r | 20 - tests/testthat/test-sample_n.r | 16 - tests/testthat/test-sas_to_csv.r | 8 - tests/testthat/test-sas_to_disk.frame.r | 10 - tests/testthat/test-semi_join.R | 72 --- tests/testthat/test-setup.r | 7 - tests/testthat/test-shard.r | 28 - tests/testthat/test-shardkey.r | 17 - tests/testthat/test-srckeep.r | 17 - tests/testthat/test-tbl_vars.r | 16 - tests/testthat/test-util.r | 8 - tests/testthat/test-write_disk.frame.R | 39 -- tests/testthat/test-zip_to_disk.frame.r | 7 - utils/build_utils.R | 20 +- 155 files changed, 2591 insertions(+), 4047 deletions(-) create mode 100644 docs/articles/01-intro_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/02-intro-disk-frame_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/03-concepts_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/04-ingesting-data_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/05-data-table-syntax_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/06-vs-dask-juliadb_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/07-glm_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/08-more-epic_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/09-convenience-features_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/10-group-by_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/11-custom-group-by_files/header-attrs-2.8/header-attrs.js create mode 100644 docs/articles/88-trouble-shooting.html create mode 100644 docs/articles/88-trouble-shooting_files/header-attrs-2.8/header-attrs.js delete mode 100644 tests/testthat.R delete mode 100644 tests/testthat/test-Rcpp.R delete mode 100644 tests/testthat/test-add-chunk.r delete mode 100644 tests/testthat/test-anti_join.R delete mode 100644 tests/testthat/test-as-data-frame.R delete mode 100644 tests/testthat/test-as-disk-frame.R delete mode 100644 tests/testthat/test-bloom-filter.r delete mode 100644 tests/testthat/test-collect.R delete mode 100644 tests/testthat/test-compute.r delete mode 100644 tests/testthat/test-csv2disk.frame.r delete mode 100644 tests/testthat/test-data-table.r delete mode 100644 tests/testthat/test-delete.r delete mode 100644 tests/testthat/test-disk-frame.r delete mode 100644 tests/testthat/test-dplyr-verbs.r delete mode 100644 tests/testthat/test-dtplyr-support.r delete mode 100644 tests/testthat/test-foverlaps.r delete mode 100644 tests/testthat/test-full_join.R delete mode 100644 tests/testthat/test-get_chunk.r delete mode 100644 tests/testthat/test-get_chunk_ids.r delete mode 100644 tests/testthat/test-glm.r delete mode 100644 tests/testthat/test-group-by.R delete mode 100644 tests/testthat/test-hard-arrange.R delete mode 100644 tests/testthat/test-inner_join.R delete mode 100644 tests/testthat/test-is-disk.frame.r delete mode 100644 tests/testthat/test-left_join.R delete mode 100644 tests/testthat/test-map.r delete mode 100644 tests/testthat/test-map2.r delete mode 100644 tests/testthat/test-merge.r delete mode 100644 tests/testthat/test-names.r delete mode 100644 tests/testthat/test-nchunks.r delete mode 100644 tests/testthat/test-nrow-ncol.R delete mode 100644 tests/testthat/test-overwrite_check.r delete mode 100644 tests/testthat/test-pls-add.r delete mode 100644 tests/testthat/test-print.r delete mode 100644 tests/testthat/test-pull.r delete mode 100644 tests/testthat/test-rbindlist.r delete mode 100644 tests/testthat/test-rechunk.r delete mode 100644 tests/testthat/test-recommend_nchunk.R delete mode 100644 tests/testthat/test-remove_chunk.r delete mode 100644 tests/testthat/test-right_join.r delete mode 100644 tests/testthat/test-sample_frac.r delete mode 100644 tests/testthat/test-sample_n.r delete mode 100644 tests/testthat/test-sas_to_csv.r delete mode 100644 tests/testthat/test-sas_to_disk.frame.r delete mode 100644 tests/testthat/test-semi_join.R delete mode 100644 tests/testthat/test-setup.r delete mode 100644 tests/testthat/test-shard.r delete mode 100644 tests/testthat/test-shardkey.r delete mode 100644 tests/testthat/test-srckeep.r delete mode 100644 tests/testthat/test-tbl_vars.r delete mode 100644 tests/testthat/test-util.r delete mode 100644 tests/testthat/test-write_disk.frame.R delete mode 100644 tests/testthat/test-zip_to_disk.frame.r diff --git a/CRAN-RELEASE b/CRAN-RELEASE index f72236de..48d603de 100644 --- a/CRAN-RELEASE +++ b/CRAN-RELEASE @@ -1,2 +1,2 @@ -This package was submitted to CRAN on 2021-02-13. -Once it is accepted, delete this file and tag the release (commit f7dd3db). +This package was submitted to CRAN on 2021-03-12. +Once it is accepted, delete this file and tag the release (commit 34bafaa). diff --git a/DESCRIPTION b/DESCRIPTION index a9207fc2..0f848ee8 100644 --- a/DESCRIPTION +++ b/DESCRIPTION @@ -1,8 +1,8 @@ Type: Package Package: disk.frame Title: Larger-than-RAM Disk-Based Data Manipulation Framework -Version: 0.4.0 -Date: 2021-02-11 +Version: 0.5.0 +Date: 2021-05-09 Authors@R: c( person("Dai", "ZJ", email = "zhuojia.dai@gmail.com", role = c("aut", "cre")), person("Jacky", "Poon", role = c("ctb")) diff --git a/NAMESPACE b/NAMESPACE index 51fa7439..da78d075 100644 --- a/NAMESPACE +++ b/NAMESPACE @@ -67,7 +67,6 @@ S3method(transmute,disk.frame) export(IQR_df.chunk_agg.disk.frame) export(IQR_df.collected_agg.disk.frame) export(add_chunk) -export(add_count.disk.frame) export(add_tally.disk.frame) export(all_df.chunk_agg.disk.frame) export(all_df.collected_agg.disk.frame) @@ -176,7 +175,6 @@ importFrom(data.table,setDT) importFrom(data.table,setkey) importFrom(data.table,setkeyv) importFrom(data.table,timetaken) -importFrom(dplyr,add_count) importFrom(dplyr,add_tally) importFrom(dplyr,anti_join) importFrom(dplyr,arrange) diff --git a/NEWS.md b/NEWS.md index f06cc6e6..088222ce 100644 --- a/NEWS.md +++ b/NEWS.md @@ -1,3 +1,9 @@ +# disk.frame 0.5 +* removed `add_count` method + +# disk.frame 0.4.1 +* removed use of `sysctl` which was violating CRAN policy + # disk.frame 0.4.0 * Removed `count` and `tally` * Fixed package compatibility diff --git a/R/dplyr_verbs.r b/R/dplyr_verbs.r index e92e678d..75be5774 100644 --- a/R/dplyr_verbs.r +++ b/R/dplyr_verbs.r @@ -70,10 +70,10 @@ chunk_arrange <- create_chunk_mapper(dplyr::arrange) # TODO alot of these .disk.frame functions are not generic -#' @export -#' @importFrom dplyr add_count -#' @rdname dplyr_verbs -add_count.disk.frame <- create_chunk_mapper(dplyr::add_count) +#' #' @export +#' #' @importFrom dplyr add_count +#' #' @rdname dplyr_verbs +#' add_count.disk.frame <- create_chunk_mapper(dplyr::add_count) #' @export diff --git a/R/recommend_nchunks.r b/R/recommend_nchunks.r index 8d77e1f5..e5be6ac0 100644 --- a/R/recommend_nchunks.r +++ b/R/recommend_nchunks.r @@ -91,16 +91,18 @@ df_ram_size <- function() { } } } else { - os = R.version$os - if (length(grep("^darwin", os))) { - a = substring(system("sysctl hw.memsize", intern = TRUE), 13) - } #else { + #os = R.version$os + #if (length(grep("^darwin", os))) { + #a = substring(system("sysctl hw.memsize", intern = TRUE), 13) + # the above is not allowed by CRAN + #} #else { # This would work but is not allowed by CRAN #a = system('grep MemTotal /proc/meminfo', intern = TRUE) #} - l = strsplit(a, " ")[[1]] - l = as.numeric(l[length(l)-1]) - ram_size = l/1024^2 + #l = strsplit(a, " ")[[1]] + #l = as.numeric(l[length(l)-1]) + #ram_size = l/1024^2 + ram_size = 16 # to be conservative } if(is.null(ram_size)) { diff --git a/README.md b/README.md index d7e33912..7f14e340 100644 --- a/README.md +++ b/README.md @@ -211,15 +211,12 @@ flights.df %>% filter(year == 2013) %>% mutate(origin_dest = paste0(origin, dest)) %>% head(2) -#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time -#> 1 2013 1 1 517 515 2 830 819 -#> 2 2013 1 1 533 529 4 850 830 -#> arr_delay carrier flight tailnum origin dest air_time distance hour minute -#> 1 11 UA 1545 N14228 EWR IAH 227 1400 5 15 -#> 2 20 UA 1714 N24211 LGA IAH 227 1416 5 29 -#> time_hour origin_dest -#> 1 2013-01-01 05:00:00 EWRIAH -#> 2 2013-01-01 05:00:00 LGAIAH +#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight +#> 1 2013 1 1 517 515 2 830 819 11 UA 1545 +#> 2 2013 1 1 533 529 4 850 830 20 UA 1714 +#> tailnum origin dest air_time distance hour minute time_hour origin_dest +#> 1 N14228 EWR IAH 227 1400 5 15 2013-01-01 05:00:00 EWRIAH +#> 2 N24211 LGA IAH 227 1416 5 29 2013-01-01 05:00:00 LGAIAH ``` ### Group-by @@ -276,14 +273,6 @@ obtained using estimated methods. ``` r library(data.table) -#> -#> Attaching package: 'data.table' -#> The following object is masked from 'package:purrr': -#> -#> transpose -#> The following objects are masked from 'package:dplyr': -#> -#> between, first, last suppressWarnings( grp_by_stage1 <- @@ -324,7 +313,7 @@ To find out where the disk.frame is stored on disk: ``` r # where is the disk.frame stored attr(flights.df, "path") -#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpMpwOj2\\file4bb025d75c63.df" +#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpEdhlRv\\file33ec67c31d9c.df" ``` A number of data.frame functions are implemented for disk.frame @@ -332,23 +321,19 @@ A number of data.frame functions are implemented for disk.frame ``` r # get first few rows head(flights.df, 1) -#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time -#> 1: 2013 1 1 517 515 2 830 819 -#> arr_delay carrier flight tailnum origin dest air_time distance hour minute -#> 1: 11 UA 1545 N14228 EWR IAH 227 1400 5 15 -#> time_hour -#> 1: 2013-01-01 05:00:00 +#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier +#> 1: 2013 1 1 517 515 2 830 819 11 UA +#> flight tailnum origin dest air_time distance hour minute time_hour +#> 1: 1545 N14228 EWR IAH 227 1400 5 15 2013-01-01 05:00:00 ``` ``` r # get last few rows tail(flights.df, 1) -#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time -#> 1: 2013 9 30 NA 840 NA NA 1020 -#> arr_delay carrier flight tailnum origin dest air_time distance hour minute -#> 1: NA MQ 3531 N839MQ LGA RDU NA 431 8 40 -#> time_hour -#> 1: 2013-09-30 08:00:00 +#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier +#> 1: 2013 9 30 NA 840 NA NA 1020 NA MQ +#> flight tailnum origin dest air_time distance hour minute time_hour +#> 1: 3531 N839MQ LGA RDU NA 431 8 40 2013-09-30 08:00:00 ``` ``` r diff --git a/book/01-intro.Rmd b/book/01-intro.Rmd index d0df5a20..0ffc5a2e 100644 --- a/book/01-intro.Rmd +++ b/book/01-intro.Rmd @@ -3,7 +3,7 @@ title: "Preface - The birth of `disk.frame`" author: "ZJ" output: rmarkdown::html_vignette vignette: > - %\VignetteIndexEntry{preface} + %\VignetteIndexEntry{Preface - The birth of `disk.frame`} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- diff --git a/book/06-vs-dask-juliadb.Rmd b/book/06-vs-dask-juliadb.Rmd index 830081f9..c018af59 100644 --- a/book/06-vs-dask-juliadb.Rmd +++ b/book/06-vs-dask-juliadb.Rmd @@ -3,7 +3,7 @@ title: "Benchmarks 1: disk.frame beats Dask! disk.frame beats JuliaDB! Anyone el author: "ZJ" output: rmarkdown::html_vignette vignette: > - %\VignetteIndexEntry{benchmark-1} + %\VignetteIndexEntry{Benchmarks 1: disk.frame beats Dask! disk.frame beats JuliaDB! Anyone else wanna challenge?} %\VignetteEngine{knitr::rmarkdown} %\VignetteEncoding{UTF-8} --- diff --git a/cran-comments.md b/cran-comments.md index 3784d366..b661d1a8 100644 --- a/cran-comments.md +++ b/cran-comments.md @@ -1,11 +1,11 @@ -## Submission for v0.4.0 -* Fixed recently reported warnings +## Submission for v0.5.0 +* Fixed issue in CRAN check but needed to update version to follow semver conventions ## Test environments -* local Windows 10 Pro install, R 4.0.3 -* local Windows 10 Pro install, R devel (as of 2021-02-11) -* local Linux/Ubuntu install, R 4.0.3 -* local Linux/Ubuntu install, R devel (as of 2021-02-11) +* local Windows 10 Pro install, R 4.0.5 +* local Windows 10 Pro install, R devel (as of 2021-05-09) +* local Linux/Ubuntu install, R 4.0.5 +* local Linux/Ubuntu install, R devel (as of 2021-05-09) ## R CMD check results There were no ERRORs nor WARNINGs nor NOTE when run locally. diff --git a/docs/404.html b/docs/404.html index 8c5ef8b1..ed268161 100644 --- a/docs/404.html +++ b/docs/404.html @@ -71,7 +71,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -79,7 +79,7 @@
  • @@ -135,7 +138,7 @@
    • - +
      • @@ -175,7 +178,7 @@

      Contents

      -

      Site built with pkgdown 1.5.1.

      +

      Site built with pkgdown 1.6.1.

      diff --git a/docs/LICENSE-text.html b/docs/LICENSE-text.html index f2435efd..a8f5414f 100644 --- a/docs/LICENSE-text.html +++ b/docs/LICENSE-text.html @@ -71,7 +71,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -79,7 +79,7 @@
    • @@ -135,7 +138,7 @@
      • - +
        • @@ -177,7 +180,7 @@

        Contents

        -

        Site built with pkgdown 1.5.1.

        +

        Site built with pkgdown 1.6.1.

        diff --git a/docs/articles/01-intro.html b/docs/articles/01-intro.html index 93f57ca8..ce4dac15 100644 --- a/docs/articles/01-intro.html +++ b/docs/articles/01-intro.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -39,7 +39,7 @@
      • @@ -95,7 +98,7 @@
        • - +
          • @@ -109,7 +112,7 @@ -
          +

          Preface - The birth of disk.frame @@ -148,7 +151,7 @@

          -

          Site built with pkgdown 1.5.1.

          +

          Site built with pkgdown 1.6.1.

          diff --git a/docs/articles/01-intro_files/header-attrs-2.8/header-attrs.js b/docs/articles/01-intro_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/01-intro_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/02-intro-disk-frame.html b/docs/articles/02-intro-disk-frame.html index 9ff4e6d6..dd9b0746 100644 --- a/docs/articles/02-intro-disk-frame.html +++ b/docs/articles/02-intro-disk-frame.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
          @@ -39,7 +39,7 @@
        • @@ -95,7 +98,7 @@
          • - +
            • @@ -109,7 +112,7 @@ -
            +

            Quick Start: Basic Operations with nycflights13

            @@ -138,19 +141,22 @@

            Installation

            Simply run

            -
            install.packages("disk.frame") # when CRAN ready
            +
            +install.packages("disk.frame") # when CRAN ready

            or

            -
            devtools::install_github("xiaodaigh/disk.frame")
            +
            +devtools::install_github("xiaodaigh/disk.frame")

            Set-up disk.frame

            disk.frame works best if it can process multiple data chunks in parallel. The best way to set-up disk.frame so that each CPU core runs a background worker is by using

            -
            setup_disk.frame()
+
            +setup_disk.frame()
 
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
            
+options(future.globals.maxSize = Inf)

            The setup_disk.frame() sets up background workers equal to the number of CPU cores; please note that, by default, hyper-threaded cores are counted as one not two.

            Alternatively, one may specify the number of workers using setup_disk.frame(workers = n).

            @@ -164,153 +170,171 @@

            Creating a disk.frame from data.frame

            We convert a data.frame to disk.frame using the as.data.frame function.

            -
            library(nycflights13)
-library(dplyr)
-library(disk.frame)
-library(data.table)
+
            +library(nycflights13)
+library(dplyr)
+library(disk.frame)
+library(data.table)
 
 # convert the flights data to a disk.frame and store the disk.frame in the folder
 # "tmp_flights" and overwrite any content if needed
-flights.df <- as.disk.frame(
-  flights,
-  outdir = file.path(tempdir(), "tmp_flights.df"),
-  overwrite = TRUE)
-flights.df
            
+flights.df <- as.disk.frame(
+  flights, 
+  outdir = file.path(tempdir(), "tmp_flights.df"),
+  overwrite = TRUE)
+flights.df

            You should now see a folder called tmp_flights with some files in it, namely 1.fst, 2.fst…. where each fst files is one chunk of the disk.frame.

            Creating a disk.frame from CSV

            -
            library(nycflights13)
+
            +library(nycflights13)
 # write a csv
-csv_path = file.path(tempdir(), "tmp_flights.csv")
-data.table::fwrite(flights, csv_path)
+csv_path = file.path(tempdir(), "tmp_flights.csv")
+data.table::fwrite(flights, csv_path)
 
 # load the csv into a disk.frame
-df_path = file.path(tempdir(), "tmp_flights.df")
-flights.df <- csv_to_disk.frame(
-  csv_path,
-  outdir = df_path,
-  overwrite = T)
-
-flights.df
            
+df_path = file.path(tempdir(), "tmp_flights.df")
+flights.df <- csv_to_disk.frame(
+  csv_path, 
+  outdir = df_path,
+  overwrite = T)
+  
+flights.df

            If the CSV is too large to read in, then we can also use the in_chunk_size option to control how many rows to read in at once. For example to read in the data 100,000 rows at a time.

            -
            library(nycflights13)
-library(disk.frame)
+
            +library(nycflights13)
+library(disk.frame)
 
 # write a csv
-csv_path = file.path(tempdir(), "tmp_flights.csv")
-
-data.table::fwrite(flights, csv_path)
+csv_path = file.path(tempdir(), "tmp_flights.csv")
 
-df_path = file.path(tempdir(), "tmp_flights.df")
+data.table::fwrite(flights, csv_path)
 
-flights.df <- csv_to_disk.frame(
-  csv_path,
-  outdir = df_path,
-  in_chunk_size = 100000)
+df_path = file.path(tempdir(), "tmp_flights.df")
 
-flights.df
            
+flights.df <- csv_to_disk.frame(
+  csv_path, 
+  outdir = df_path, 
+  in_chunk_size = 100000)
+#> Warning: UNRELIABLE VALUE: Future ('<none>') unexpectedly generated random
+#> numbers without specifying argument 'seed'. There is a risk that those random
+#> numbers are not statistically sound and the overall results might be invalid.
+#> To fix this, specify 'seed=TRUE'. This ensures that proper, parallel-safe random
+#> numbers are produced via the L'Ecuyer-CMRG method. To disable this check, use
+#> 'seed=NULL', or set option 'future.rng.onMisuse' to "ignore".
+  
+flights.df

            disk.frame also has a function zip_to_disk.frame that can convert every CSV in a zip file to disk.frames.

            Simple dplyr verbs and lazy evaluation

            -
            flights.df1 <- select(flights.df, year:day, arr_delay, dep_delay)
-flights.df1
            -
            class(flights.df1)
-#> [1] "disk.frame"        "disk.frame.folder"
            +
            +flights.df1 <- select(flights.df, year:day, arr_delay, dep_delay)
+flights.df1
            +
            +class(flights.df1)
+#> [1] "disk.frame"        "disk.frame.folder"

            The class of flights.df1 is also a disk.frame after the dplyr::select transformation. Also, disk.frame operations are by default (and where possible) lazy, meaning it doesn’t perform the operations right away. Instead, it waits until you call collect. Exceptions to this rule are the *_join operations which evaluated eagerly under certain conditions see Joins for disk.frame in-depth for details.

            For lazily constructed disk.frames (e.g. flights.df1). The function collect can be used to bring the results from disk into R, e.g.

            -
            collect(flights.df1) %>% head(2)
+
            +collect(flights.df1) %>% head(2)
 #>    year month day arr_delay dep_delay
 #> 1: 2013     1   1        11         2
-#> 2: 2013     1   1        20         4
            
+#> 2: 2013     1   1        20         4

            Of course, for larger-than-RAM datasets, one wouldn’t call collect on the whole disk.frame (because why would you need disk.frame otherwise). More likely, one would call collect on a filtered dataset or one summarized with group_by.

            Some examples of other dplyr verbs applied:

            -
            filter(flights.df, dep_delay > 1000) %>% collect %>% head(2)
-#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
-#> 1 2013     1   9      641            900      1301     1242           1530
-#> 2 2013     1  10     1121           1635      1126     1239           1810
-#>   arr_delay carrier flight tailnum origin dest air_time distance hour minute
-#> 1      1272      HA     51  N384HA    JFK  HNL      640     4983    9      0
-#> 2      1109      MQ   3695  N517MQ    EWR  ORD      111      719   16     35
+
            +filter(flights.df, dep_delay > 1000) %>% collect %>% head(2)
+#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
+#> 1: 2013     1   9      641            900      1301     1242           1530
+#> 2: 2013     1  10     1121           1635      1126     1239           1810
+#>    arr_delay carrier flight tailnum origin dest air_time distance hour minute
+#> 1:      1272      HA     51  N384HA    JFK  HNL      640     4983    9      0
+#> 2:      1109      MQ   3695  N517MQ    EWR  ORD      111      719   16     35
 #>              time_hour
-#> 1 2013-01-09T14:00:00Z
-#> 2 2013-01-10T21:00:00Z
            
-
            mutate(flights.df, speed = distance / air_time * 60) %>% collect %>% head(2)
-#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
-#> 1 2013     1   1      517            515         2      830            819
-#> 2 2013     1   1      533            529         4      850            830
-#>   arr_delay carrier flight tailnum origin dest air_time distance hour minute
-#> 1        11      UA   1545  N14228    EWR  IAH      227     1400    5     15
-#> 2        20      UA   1714  N24211    LGA  IAH      227     1416    5     29
+#> 1: 2013-01-09 14:00:00
+#> 2: 2013-01-10 21:00:00
            
+
            +mutate(flights.df, speed = distance / air_time * 60) %>% collect %>% head(2)
+#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
+#> 1: 2013     1   1      517            515         2      830            819
+#> 2: 2013     1   1      533            529         4      850            830
+#>    arr_delay carrier flight tailnum origin dest air_time distance hour minute
+#> 1:        11      UA   1545  N14228    EWR  IAH      227     1400    5     15
+#> 2:        20      UA   1714  N24211    LGA  IAH      227     1416    5     29
 #>              time_hour    speed
-#> 1 2013-01-01T10:00:00Z 370.0441
-#> 2 2013-01-01T10:00:00Z 374.2731
            
+#> 1: 2013-01-01 10:00:00 370.0441
+#> 2: 2013-01-01 10:00:00 374.2731

            Examples of NOT fully supported dplyr verbs

            The chunk_arrange function arranges (sorts) each chunk but not the whole dataset. So use with caution. Similarly chunk_summarise creates summary variables within each chunk and hence also needs to be used with caution. In the Group By section, we demonstrate how to use summarise in the disk.frame context correctly with hard_group_bys.

            -
            # this only sorts within each chunk
-chunk_arrange(flights.df, dplyr::desc(dep_delay)) %>% collect %>% head(2)
-#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
-#> 1 2013     1   9      641            900      1301     1242           1530
-#> 2 2013     1  10     1121           1635      1126     1239           1810
-#>   arr_delay carrier flight tailnum origin dest air_time distance hour minute
-#> 1      1272      HA     51  N384HA    JFK  HNL      640     4983    9      0
-#> 2      1109      MQ   3695  N517MQ    EWR  ORD      111      719   16     35
+
            +# this only sorts within each chunk
+chunk_arrange(flights.df, dplyr::desc(dep_delay)) %>% collect %>% head(2)
+#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
+#> 1: 2013     1   9      641            900      1301     1242           1530
+#> 2: 2013     1  10     1121           1635      1126     1239           1810
+#>    arr_delay carrier flight tailnum origin dest air_time distance hour minute
+#> 1:      1272      HA     51  N384HA    JFK  HNL      640     4983    9      0
+#> 2:      1109      MQ   3695  N517MQ    EWR  ORD      111      719   16     35
 #>              time_hour
-#> 1 2013-01-09T14:00:00Z
-#> 2 2013-01-10T21:00:00Z
            
-
            chunk_summarize(flights.df, mean_dep_delay = mean(dep_delay, na.rm =T)) %>% collect
-#>   mean_dep_delay
-#> 1       12.32700
-#> 2       12.01291
-#> 3       13.81315
-#> 4       12.94445
-#> 5       12.67813
-#> 6       12.05854
            
+#> 1: 2013-01-09 14:00:00
+#> 2: 2013-01-10 21:00:00
            +
            +chunk_summarize(flights.df, mean_dep_delay = mean(dep_delay, na.rm =T)) %>% collect
+#>    mean_dep_delay
+#> 1:       12.32700
+#> 2:       12.01291
+#> 3:       13.81315
+#> 4:       12.94445
+#> 5:       12.67813
+#> 6:       12.05854

            Piping

            One can chain dplyr verbs together like with a data.frame

            -
            c4 <- flights %>%
-  filter(month == 5, day == 17, carrier %in% c('UA', 'WN', 'AA', 'DL')) %>%
-  select(carrier, dep_delay, air_time, distance) %>%
-  mutate(air_time_hours = air_time / 60) %>%
-  collect %>%
-  arrange(carrier)# arrange should occur after `collect`
+
            +c4 <- flights %>%
+  filter(month == 5, day == 17, carrier %in% c('UA', 'WN', 'AA', 'DL')) %>%
+  select(carrier, dep_delay, air_time, distance) %>%
+  mutate(air_time_hours = air_time / 60) %>%
+  collect %>%
+  arrange(carrier)# arrange should occur after `collect`
 
-c4  %>% head
+c4  %>% head
 #>    carrier dep_delay air_time distance air_time_hours
 #> 1:      AA        -7      142     1089       2.366667
 #> 2:      AA        -9      186     1389       3.100000
 #> 3:      AA        -6      143     1096       2.383333
 #> 4:      AA        -4      114      733       1.900000
 #> 5:      AA        -2      146     1085       2.433333
-#> 6:      AA        -7      119      733       1.983333
            
+#> 6:      AA        -7      119      733       1.983333

            List of supported dplyr verbs

            -
            select
-rename
-filter
-chunk_arrange # within each chunk
-chunk_group_by # within each chunk
-chunk_summarize # within each chunk
-group_by # limited functions
-summarize # limited functions
-mutate
-transmute
-left_join
-inner_join
-full_join # careful. Performance!
-semi_join
-anit_join
            +
            +select
+rename
+filter
+chunk_arrange # within each chunk
+chunk_group_by # within each chunk
+chunk_summarize # within each chunk
+group_by # limited functions
+summarize # limited functions
+mutate
+transmute
+left_join
+inner_join
+full_join # careful. Performance!
+semi_join
+anit_join
            @@ -322,12 +346,13 @@

            Group-by

            -

            {disk.frame} implements the group_by operation some caveats. In the {disk.frame} framework, only a set functions are supported in summarize. However, the user can create more custom group-by functions can be defined.

            -
            flights.df %>%
-  group_by(carrier) %>% # notice that hard_group_by needs to be set
-  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
-  collect %>%
-  arrange(carrier)
+
            disk.frame implements the group_by operation some caveats. In the disk.frame framework, only a set functions are supported in summarize. However, the user can create more custom group-by functions can be defined.
            
+
            +flights.df %>%
+  group_by(carrier) %>% # notice that hard_group_by needs to be set
+  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
+  collect %>% 
+  arrange(carrier)
 #> # A tibble: 16 x 3
 #>    carrier count mean_dep_delay
 #>    <chr>   <int>          <dbl>
@@ -346,17 +371,18 @@ 
            
 #> 13 US      20536           3.78
 #> 14 VX       5162          12.9 
 #> 15 WN      12275          17.7 
-#> 16 YV        601          19.0
            
+#> 16 YV        601          19.0

            Restrict input columns for faster processing

            One can restrict which input columns to load into memory for each chunk; this can significantly increase the speed of data processing. To restrict the input columns, use the srckeep function which only accepts column names as a string vector.

            -
            flights.df %>%
-  srckeep(c("carrier","dep_delay")) %>%
-  group_by(carrier) %>%
-  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
-  collect
+
            +flights.df %>%
+  srckeep(c("carrier","dep_delay")) %>%
+  group_by(carrier) %>% 
+  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
+  collect
 #> # A tibble: 16 x 3
 #>    carrier count mean_dep_delay
 #>    <chr>   <int>          <dbl>
@@ -375,29 +401,31 @@ 
            
 #> 13 US      20536           3.78
 #> 14 VX       5162          12.9 
 #> 15 WN      12275          17.7 
-#> 16 YV        601          19.0
            
+#> 16 YV        601          19.0

            Input column restriction is one of the most critical efficiencies provided by disk.frame. Because the underlying format allows random access to columns (i.e. retrieve only the columns used for processing), hence one can drastically reduce the amount of data loaded into RAM for processing by keeping only those columns that are directly used to produce the results.

            Joins

            disk.frame supports many dplyr joins including:

            -
            left_join
-inner_join
-semi_join
-inner_join
-full_join # requires hard_group_by on both left and right
            +
            +left_join
+inner_join
+semi_join
+inner_join
+full_join # requires hard_group_by on both left and right

            In all cases, the left dataset (x) must be a disk.frame, and the right dataset (y) can be either a disk.frame or a data.frame. If the right dataset is a disk.frame and the shardkeys are different between the two disk.frames then two expensive hard group_by operations are performed eagerly, one on the left disk.frame and one on the right disk.frame to perform the joins correctly.

            However, if the right dataset is a data.frame then hard_group_bys are only performed in the case of full_join.

            Note disk.frame does not support right_join the user should use left_join instead.

            The below joins are performed lazily because airlines.dt is a data.table not a disk.frame:

            -
            # make airlines a data.table
-airlines.dt <- data.table(airlines)
+
            +# make airlines a data.table
+airlines.dt <- data.table(airlines)
 # flights %>% left_join(airlines, by = "carrier") #
-flights.df %>%
-  left_join(airlines.dt, by ="carrier") %>%
-  collect %>%
-  head
+flights.df %>% 
+  left_join(airlines.dt, by ="carrier") %>% 
+  collect %>% 
+  head
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   1      517            515         2      830            819
 #> 2: 2013     1   1      533            529         4      850            830
@@ -412,17 +440,18 @@ 
            
 #> 4:       -18      B6    725  N804JB    JFK  BQN      183     1576    5     45
 #> 5:       -25      DL    461  N668DN    LGA  ATL      116      762    6      0
 #> 6:        12      UA   1696  N39463    EWR  ORD      150      719    5     58
-#>               time_hour                   name
-#> 1: 2013-01-01T10:00:00Z  United Air Lines Inc.
-#> 2: 2013-01-01T10:00:00Z  United Air Lines Inc.
-#> 3: 2013-01-01T10:00:00Z American Airlines Inc.
-#> 4: 2013-01-01T10:00:00Z        JetBlue Airways
-#> 5: 2013-01-01T11:00:00Z   Delta Air Lines Inc.
-#> 6: 2013-01-01T10:00:00Z  United Air Lines Inc.
            
-
            flights.df %>%
-  left_join(airlines.dt, by = c("carrier")) %>%
-  collect %>%
-  tail
+#>              time_hour                   name
+#> 1: 2013-01-01 10:00:00  United Air Lines Inc.
+#> 2: 2013-01-01 10:00:00  United Air Lines Inc.
+#> 3: 2013-01-01 10:00:00 American Airlines Inc.
+#> 4: 2013-01-01 10:00:00        JetBlue Airways
+#> 5: 2013-01-01 11:00:00   Delta Air Lines Inc.
+#> 6: 2013-01-01 10:00:00  United Air Lines Inc.
            
+
            +flights.df %>% 
+  left_join(airlines.dt, by = c("carrier")) %>% 
+  collect %>% 
+  tail
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     9  30       NA           1842        NA       NA           2019
 #> 2: 2013     9  30       NA           1455        NA       NA           1634
@@ -437,82 +466,77 @@ 
            
 #> 4:        NA      MQ   3461  N535MQ    LGA  BNA       NA      764   12     10
 #> 5:        NA      MQ   3572  N511MQ    LGA  CLE       NA      419   11     59
 #> 6:        NA      MQ   3531  N839MQ    LGA  RDU       NA      431    8     40
-#>               time_hour                     name
-#> 1: 2013-09-30T22:00:00Z ExpressJet Airlines Inc.
-#> 2: 2013-09-30T18:00:00Z        Endeavor Air Inc.
-#> 3: 2013-10-01T02:00:00Z        Endeavor Air Inc.
-#> 4: 2013-09-30T16:00:00Z                Envoy Air
-#> 5: 2013-09-30T15:00:00Z                Envoy Air
-#> 6: 2013-09-30T12:00:00Z                Envoy Air
            
+#>              time_hour                     name
+#> 1: 2013-09-30 22:00:00 ExpressJet Airlines Inc.
+#> 2: 2013-09-30 18:00:00        Endeavor Air Inc.
+#> 3: 2013-10-01 02:00:00        Endeavor Air Inc.
+#> 4: 2013-09-30 16:00:00                Envoy Air
+#> 5: 2013-09-30 15:00:00                Envoy Air
+#> 6: 2013-09-30 12:00:00                Envoy Air

            Window functions and arbitrary functions

            -

            {disk.frame} supports all data.frame operations, unlike Spark which can only perform those operations that Spark has implemented. Hence windowing functions like min_rank and rank are supported out of the box.

            +

            disk.frame supports all data.frame operations, unlike Spark which can only perform those operations that Spark has implemented. Hence windowing functions like min_rank and rank are supported out of the box.

            For the following example, we will use the hard_group_by which performs a group-by and also reorganises the chunks so that all records with the same year, month, and day end up in the same chunk. This is typically not advised, as hard_group_by can be slow for large datasets.

            -
            # Find the most and least delayed flight each day
-bestworst <- flights.df %>%
-   srckeep(c("year","month","day", "dep_delay")) %>%
-   hard_group_by(c("year", "month", "day")) %>%
-   filter(dep_delay == min(dep_delay, na.rm = T) || dep_delay == max(dep_delay, na.rm = T)) %>%
-   collect
-
-bestworst %>% head
-#> # A tibble: 6 x 4
-#> # Groups:   year, month, day [1]
-#>    year month   day dep_delay
-#>   <int> <int> <int>     <int>
-#> 1  2013     2    21       301
-#> 2  2013     2    21        -9
-#> 3  2013     2    21        -1
-#> 4  2013     2    21         2
-#> 5  2013     2    21        -4
-#> 6  2013     2    21        10
            +
            +# Find the most and least delayed flight each day
+bestworst <- flights.df %>%
+   srckeep(c("year","month","day", "dep_delay")) %>%
+   hard_group_by(c("year", "month", "day")) %>%
+   filter(dep_delay == min(dep_delay, na.rm = T) || dep_delay == max(dep_delay, na.rm = T)) %>%
+   collect
+   
+bestworst %>% head
+#>    year month day dep_delay
+#> 1: 2013     2  21       301
+#> 2: 2013     2  21        -9
+#> 3: 2013     2  21        -1
+#> 4: 2013     2  21         2
+#> 5: 2013     2  21        -4
+#> 6: 2013     2  21        10

            another example

            -
            ranked <- flights.df %>%
-  srckeep(c("year","month","day", "dep_delay")) %>%
-  hard_group_by(c("year", "month", "day")) %>%
-  filter(min_rank(desc(dep_delay)) <= 2 & dep_delay > 0) %>%
-  collect
+
            +ranked <- flights.df %>%
+  srckeep(c("year","month","day", "dep_delay")) %>%
+  hard_group_by(c("year", "month", "day")) %>%
+  filter(min_rank(desc(dep_delay)) <= 2 & dep_delay > 0) %>%
+  collect
 
-ranked %>% head
-#> # A tibble: 6 x 4
-#> # Groups:   year, month, day [3]
-#>    year month   day dep_delay
-#>   <int> <int> <int>     <int>
-#> 1  2013     1     9      1301
-#> 2  2013     1     9       253
-#> 3  2013     1    10      1126
-#> 4  2013     1    10       385
-#> 5  2013     1    17       259
-#> 6  2013     1    17       255
            
+ranked %>% head
+#>    year month day dep_delay
+#> 1: 2013     1   9      1301
+#> 2: 2013     1   9       253
+#> 3: 2013     1  10      1126
+#> 4: 2013     1  10       385
+#> 5: 2013     1  17       259
+#> 6: 2013     1  17       255

            one more example

            -
            # Rank each flight within a daily
-ranked <- flights.df %>%
-  srckeep(c("year","month","day", "dep_delay")) %>%
-  chunk_group_by(year, month, day) %>%
-  select(dep_delay) %>%
-  mutate(rank = rank(desc(dep_delay))) %>%
-  collect
+
            +# Rank each flight within a daily
+ranked <- flights.df %>%
+  srckeep(c("year","month","day", "dep_delay")) %>%
+  chunk_group_by(year, month, day) %>%
+  select(dep_delay) %>%
+  mutate(rank = rank(desc(dep_delay))) %>%
+  collect
 
-ranked %>% head
-#> # A tibble: 6 x 5
-#> # Groups:   year, month, day [1]
-#>    year month   day dep_delay  rank
-#>   <int> <int> <int>     <int> <dbl>
-#> 1  2013     1     1         2   313
-#> 2  2013     1     1         4   276
-#> 3  2013     1     1         2   313
-#> 4  2013     1     1        -1   440
-#> 5  2013     1     1        -6   742
-#> 6  2013     1     1        -4   633
            
+ranked %>% head
+#>    year month day dep_delay rank
+#> 1: 2013     1   1         2  313
+#> 2: 2013     1   1         4  276
+#> 3: 2013     1   1         2  313
+#> 4: 2013     1   1        -1  440
+#> 5: 2013     1   1        -6  742
+#> 6: 2013     1   1        -4  633

            Arbitrary by-chunk processing

            One can apply arbitrary transformations to each chunk of the disk.frame by using the delayed function which evaluates lazily or the map.disk.frame(lazy = F) function which evaluates eagerly. For example to return the number of rows in each chunk

            -
            flights.df1 <- delayed(flights.df, ~nrow(.x))
-collect_list(flights.df1) %>% head # returns number of rows for each data.frame in a list
+
            +flights.df1 <- delayed(flights.df, ~nrow(.x))
+collect_list(flights.df1) %>% head # returns number of rows for each data.frame in a list
 #> [[1]]
 #> [1] 56131
 #> 
@@ -529,9 +553,10 @@ 
            
 #> [1] 56131
 #> 
 #> [[6]]
-#> [1] 56121
            
+#> [1] 56121

            and to do the same with map.disk.frame

            -
            map(flights.df, ~nrow(.x), lazy = F) %>% head
+
            +map(flights.df, ~nrow(.x), lazy = F) %>% head
 #> Warning in map.disk.frame(flights.df, ~nrow(.x), lazy = F): map(df, ...) where
 #> df is a disk.frame has been deprecated. Please use cmap(df,...) instead
 #> [[1]]
@@ -550,15 +575,16 @@ 
            
 #> [1] 56131
 #> 
 #> [[6]]
-#> [1] 56121
            
+#> [1] 56121

            The map function can also output the results to another disk.frame folder, e.g.

            -
            # return the first 10 rows of each chunk
-flights.df2 <- map(flights.df, ~.x[1:10,], lazy = F, outdir = file.path(tempdir(), "tmp2"), overwrite = T)
+
            +# return the first 10 rows of each chunk
+flights.df2 <- map(flights.df, ~.x[1:10,], lazy = F, outdir = file.path(tempdir(), "tmp2"), overwrite = T)
 #> Warning in map.disk.frame(flights.df, ~.x[1:10, ], lazy = F, outdir =
 #> file.path(tempdir(), : map(df, ...) where df is a disk.frame has been
 #> deprecated. Please use cmap(df,...) instead
 
-flights.df2 %>% head
+flights.df2 %>% head
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   1      517            515         2      830            819
 #> 2: 2013     1   1      533            529         4      850            830
@@ -573,47 +599,49 @@ 
            
 #> 4:       -18      B6    725  N804JB    JFK  BQN      183     1576    5     45
 #> 5:       -25      DL    461  N668DN    LGA  ATL      116      762    6      0
 #> 6:        12      UA   1696  N39463    EWR  ORD      150      719    5     58
-#>               time_hour
-#> 1: 2013-01-01T10:00:00Z
-#> 2: 2013-01-01T10:00:00Z
-#> 3: 2013-01-01T10:00:00Z
-#> 4: 2013-01-01T10:00:00Z
-#> 5: 2013-01-01T11:00:00Z
-#> 6: 2013-01-01T10:00:00Z
            
-
            Notice {disk.frame} supports the purrr syntax for defining a function using ~.
            
+#>              time_hour
+#> 1: 2013-01-01 10:00:00
+#> 2: 2013-01-01 10:00:00
+#> 3: 2013-01-01 10:00:00
+#> 4: 2013-01-01 10:00:00
+#> 5: 2013-01-01 11:00:00
+#> 6: 2013-01-01 10:00:00
            +

            Notice disk.frame supports the purrr syntax for defining a function using ~.

            Sampling

            In the disk.frame framework, sampling a proportion of rows within each chunk can be performed using sample_frac.

            -
            flights.df %>% sample_frac(0.01) %>% collect %>% head
-#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
-#> 1 2013     5  20      933            935        -2     1152           1208
-#> 2 2013     5  17     1624           1622         2     1726           1735
-#> 3 2013     5  21      743            745        -2      955            956
-#> 4 2013     2   1      553            600        -7      821            825
-#> 5 2013     5  22      757            759        -2     1015           1028
-#> 6 2013     5  16     1726           1725         1     2027           2020
-#>   arr_delay carrier flight tailnum origin dest air_time distance hour minute
-#> 1       -16      EV   4140  N14543    EWR  ATL      107      746    9     35
-#> 2        -9      EV   4299  N13994    EWR  DCA       38      199   16     22
-#> 3        -1      EV   4237  N15985    EWR  CHS       99      628    7     45
-#> 4        -4      MQ   4650  N1EAMQ    LGA  ATL      125      762    6      0
-#> 5       -13      DL   2047  N662DN    LGA  ATL      104      762    7     59
-#> 6         7      AA   1901  N3FAAA    JFK  IAH      204     1417   17     25
+
            +flights.df %>% sample_frac(0.01) %>% collect %>% head
+#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
+#> 1: 2013     1  16     1700           1708        -8     2018           2023
+#> 2: 2013     1  13     1115           1115         0     1407           1425
+#> 3: 2013     2   5     2005           1935        30       NA           2220
+#> 4: 2013     1   3     1425           1429        -4     1711           1754
+#> 5: 2013     5  19     1453           1455        -2     1638           1638
+#> 6: 2013    12  20     1618           1600        18     1745           1719
+#>    arr_delay carrier flight tailnum origin dest air_time distance hour minute
+#> 1:        -5      B6     15  N334JB    JFK  FLL      175     1069   17      8
+#> 2:       -18      AA   2099  N3AWAA    LGA  MIA      150     1096   11     15
+#> 3:        NA      EV   4204  N17196    EWR  OKC       NA     1325   19     35
+#> 4:       -43      UA    257  N508UA    JFK  SFO      326     2586   14     29
+#> 5:         0      9E   3318  N908XJ    JFK  BUF       50      301   14     55
+#> 6:        26      US   2187  N946UW    LGA  DCA       46      214   16      0
 #>              time_hour
-#> 1 2013-05-20T13:00:00Z
-#> 2 2013-05-17T20:00:00Z
-#> 3 2013-05-21T11:00:00Z
-#> 4 2013-02-01T11:00:00Z
-#> 5 2013-05-22T11:00:00Z
-#> 6 2013-05-16T21:00:00Z
            
+#> 1: 2013-01-16 22:00:00
+#> 2: 2013-01-13 16:00:00
+#> 3: 2013-02-06 00:00:00
+#> 4: 2013-01-03 19:00:00
+#> 5: 2013-05-19 18:00:00
+#> 6: 2013-12-20 21:00:00

            Writing Data

            One can output a disk.frame by using the write_disk.frame function. E.g.

            -
            write_disk.frame(flights.df, outdir="out")
            +
            +write_disk.frame(flights.df, outdir="out")

            this will output a disk.frame to the folder “out”

            @@ -634,7 +662,7 @@

            -

            Site built with pkgdown 1.5.1.

            +

            Site built with pkgdown 1.6.1.

            diff --git a/docs/articles/02-intro-disk-frame_files/header-attrs-2.8/header-attrs.js b/docs/articles/02-intro-disk-frame_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/02-intro-disk-frame_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/03-concepts.html b/docs/articles/03-concepts.html index 639f53d2..a238713e 100644 --- a/docs/articles/03-concepts.html +++ b/docs/articles/03-concepts.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
          • @@ -39,7 +39,7 @@
        • @@ -95,7 +98,7 @@
          • - +
            • @@ -109,7 +112,7 @@ -
            +

            Key {disk.frame} concepts

            @@ -135,9 +138,10 @@

            Workers and parallelism

            -

            Parallelism in {disk.frame} is achieved using the {future} package. When performing many tasks, {disk.frame} uses multiple workers to perform the tasks in parallel. A worker is an R session.

            -

            It is recommended that you run the following immediately after library(disk.frame) to set-up multiple workers. For example:

            -
            library(disk.frame)
+
            Parallelism in {disk.frame} is achieved using the {future} package. When performing many tasks, disk.frame uses multiple workers to perform the tasks in parallel. A worker is an R session.
            
+
            It is recommended that you run the following immediately after library(disk.frame) to set-up multiple workers. For example:
            
+
            +library(disk.frame)
 #> Loading required package: dplyr
 #> 
 #> Attaching package: 'dplyr'
@@ -174,20 +178,23 @@ 
            
 #> The following objects are masked from 'package:base':
 #> 
 #>     colnames, ncol, nrow
-setup_disk.frame()
+setup_disk.frame()
 #> The number of workers available for disk.frame is 6
 
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
            
+options(future.globals.maxSize = Inf)

            For example, suppose we wish to compute the number of rows for each chunk, we can clearly perform this simultaneously in parallel. The code to do that is

            -
            # use only one column is fastest
-df[,.N, keep = "first_col"]
            +
            +# use only one column is fastest
+df[,.N, keep = "first_col"]

            or equivalent using the srckeep function

            -
            # use only one column is fastest
-srckeep(df, "first_col")[,.N, keep = "first_col"]
            +
            +# use only one column is fastest
+srckeep(df, "first_col")[,.N, keep = "first_col"]

            Say there are n chunks in df, and there are m workers. Then the first m chunks will run chunk[,.N] simultaneously.

            -

            To see how many workers are available for {disk.frame} to use, run

            -
            future::nbrOfWorkers()
            +

            To see how many workers are available for disk.frame to use, run

            +
            +future::nbrOfWorkers()

            @@ -214,7 +221,7 @@

            -

            Site built with pkgdown 1.5.1.

            +

            Site built with pkgdown 1.6.1.

            diff --git a/docs/articles/03-concepts_files/header-attrs-2.8/header-attrs.js b/docs/articles/03-concepts_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/03-concepts_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/04-ingesting-data.html b/docs/articles/04-ingesting-data.html index 28168dd7..f1abeeb6 100644 --- a/docs/articles/04-ingesting-data.html +++ b/docs/articles/04-ingesting-data.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0

            @@ -39,7 +39,7 @@
          • @@ -95,7 +98,7 @@
            • - +
              • @@ -109,7 +112,7 @@ -
              +

              Ingesting Data

              @@ -126,29 +129,33 @@

              ZJ

              Ingesting Data

              -

              One of the most important tasks to perform before using the {disk.frame} package is to make some disk.frames! There are a few functions to help you do that. Before we do that, we set up the {disk.frame} as usual

              +

              One of the most important tasks to perform before using the disk.frame package is to make some disk.frames! There are a few functions to help you do that. Before we do that, we set up the disk.frame as usual

              Setting up

              -
              library(disk.frame)
+
              +library(disk.frame)
 
 # set-up disk.frame to use multiple workers
-if(interactive()) {
-  setup_disk.frame()
+if(interactive()) {
+  setup_disk.frame()
   # highly recommended, however it is pun into interactive() for CRAN because
   # change user options are not allowed on CRAN
-  options(future.globals.maxSize = Inf)
-} else {
-  setup_disk.frame(2)
-}
              
+  options(future.globals.maxSize = Inf)  
+} else {
+  setup_disk.frame(2)
+}

              Convert a data.frame to disk.frame

              Firstly, there is as.disk.frame() which allows you to make a disk.frame from a data.frame, e.g.

              -
              flights.df = as.disk.frame(nycflights13::flights)
              +
              +flights.df = as.disk.frame(nycflights13::flights)

              will convert the nycflights13::flights data.frame to a disk.frame somewhere in tempdir(). To find out the location of the disk.frame use:

              -
              attr(flights.df, "path")
              +
              +attr(flights.df, "path")

              You can also specify a location to output the disk.frame to using outdir

              -
              flights.df = as.disk.frame(nycflights13::flights, outdir = "some/path.df")
              +
              +flights.df = as.disk.frame(nycflights13::flights, outdir = "some/path.df")

              it is recommended that you use .df as the extension for a disk.frame, however this is not an enforced requirement.

              However, one of the reasons for disk.frame to exist is to handle larger-than-RAM files, hence as.disk.frame is not all that useful because it can only convert data that can fit into RAM. disk.frame comes with a couple more ways to create disk.frame.

              @@ -156,66 +163,73 @@

              Creating disk.frame from CSVs

              The function csv_to_disk.frame can convert CSV files to disk.frame. The most basic usage is

              -
              some.df = csv_to_disk.frame("some/path.csv", outdir = "some.df")
              +
              +some.df = csv_to_disk.frame("some/path.csv", outdir = "some.df")

              this will convert the CSV file "some/path.csv" to a disk.frame.

              Multiple CSV files

              However, sometimes we have multiple CSV files that you want to read in and row-bind into one large disk.frame. You can do so by supplying a vector of file paths e.g. from the result of list.files

              -
              some.df = csv_to_disk.frame(c("some/path/file1.csv", "some/path/file2.csv"))
+
              +some.df = csv_to_disk.frame(c("some/path/file1.csv", "some/path/file2.csv"))
 
 # or
-some.df = csv_to_disk.frame(list.files("some/path"))
              
+some.df = csv_to_disk.frame(list.files("some/path"))

              Ingesting CSV files chunk-wise

              The csv_to_disk.frame(path, ...) function reads the file located at path in full into RAM but sometimes the CSV file may be too large to read in one go, as that would require loading the whole file into RAM. In that case, you can read the files chunk-by-chunk by using the in_chunk_size argument which controls how many rows you read in per chunk

              -
              # to read in 1 million (=1e6) rows per chunk
-csv_to_disk.frame(path, in_chunk_size = 1e6)
              +
              +# to read in 1 million (=1e6) rows per chunk
+csv_to_disk.frame(path, in_chunk_size = 1e6)

              When in_chunk_size is specified, the input file is split into many smaller files using bigreadr’s split file functions. This is generally the fastest way to ingest large CSVs, as the split files can be processed in parallel using all CPU cores. But the disk space requirement is doubled because the split files are as large as the original file. If you run out of disk space, then you must clean R’s temporary folder at tempdir() and choose another chunk_reader e.g. csv_to_disk.frame(..., chunk_reader = "LaF").

              Sharding

              One of the most important aspects of disk.frame is sharding. One can shard a disk.frame at read time by using the shardby

              -
              csv_to_disk.frame(path, shardby = "id")
              +
              +csv_to_disk.frame(path, shardby = "id")

              In the above case, all rows with the same id values will end up in the same chunk.

              Just-in-time transformation

              Sometimes, one may wish to perform some transformation on the CSV before writing out to disk. One can use the inmapfn argument to do that. The inmapfn name comes from INput MAPping FuNction. The general usage pattern is as follows:

              -
              csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
-  some_transformation(chunk)
-})
              +
              +csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
+  some_transformation(chunk)
+})

              As a contrived example, suppose you wish to convert a string into date at read time:

              -
              df = data.frame(date_str = c("2019-01-02", "2019-01-02"))
+
              +df = data.frame(date_str = c("2019-01-02", "2019-01-02"))
 
 # write the data.frame 
-write.csv(df, file.path(tempdir(), "df.csv"))
+write.csv(df, file.path(tempdir(), "df.csv"))
 
 
 # this would show that date_str is a string
-str(collect(csv_to_disk.frame(file.path(tempdir(), "df.csv")))$date_str)
+str(collect(csv_to_disk.frame(file.path(tempdir(), "df.csv")))$date_str)
 ## chr [1:2] "2019-01-02" "2019-01-02"
 
 # this would show that date_str is a string
-df = csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
+df = csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
   # convert to date_str to date format and store as "date"
-  chunk[, date := as.Date(date_str, "%Y-%m-%d")]
-  chunk[, date_str:=NULL]
-})
+  chunk[, date := as.Date(date_str, "%Y-%m-%d")]
+  chunk[, date_str:=NULL]
+})
 
-str(collect(df)$date)
-## Date[1:2], format: "2019-01-02" "2019-01-02"
              
+str(collect(df)$date)
+## Date[1:2], format: "2019-01-02" "2019-01-02"

              Reading CSVs from zip files

              Often, CSV comes zipped in a zip files. You can use the zip_to_disk.frame to convert all CSVs within a zip file

              -
              zip_to_disk.frame(path_to_zip_file)
              +
              +zip_to_disk.frame(path_to_zip_file)

              The arguments for zip_to_disk.frame are the same as csv_to_disk.frame’s.

              @@ -223,20 +237,22 @@

              Using add_chunk

              What if the method of converting to a disk.frame isn’t implemented in disk.frame yet? One can use some lower level constructs provided by disk.frame to create disk.frames. For example, the add_chunk function can be used to add more chunks to a disk.frame, e.g.

              -
              a.df = disk.frame() # create an empty disk.frame
-add_chunk(a.df, cars) # adds cars as chunk 1
-add_chunk(a.df, cars) # adds cars as chunk 2
              +
              +a.df = disk.frame() # create an empty disk.frame
+add_chunk(a.df, cars) # adds cars as chunk 1
+add_chunk(a.df, cars) # adds cars as chunk 2

              Another example of using add_chunk is via readr’s chunked read functions to create a delimited file reader

              -
              delimited_to_disk.frame <- function(file, outdir, ...) {
-  res.df = disk.frame(outdir, ...)
-  readr::read_delim_chunked(file, callback = function(chunk) {
-    add_chunk(res.df, chunk)
-  }, ...)
-
-  res.df
-}
+
              +delimited_to_disk.frame <- function(file, outdir, ...) {
+  res.df = disk.frame(outdir, ...)
+  readr::read_delim_chunked(file, callback = function(chunk) {
+    add_chunk(res.df, chunk)
+  }, ...)
+  
+  res.df
+}
 
-delimited_to_disk.frame(path, outdir = "some.df")
              
+delimited_to_disk.frame(path, outdir = "some.df")

              The above code uses readr’s read_delim_chunked function to read file and call add_chunk. The problem with this approach is that is it sequential in nature and hence is not able to take advantage of parallelism.

              @@ -262,7 +278,7 @@

              -

              Site built with pkgdown 1.5.1.

              +

              Site built with pkgdown 1.6.1.

              diff --git a/docs/articles/04-ingesting-data_files/header-attrs-2.8/header-attrs.js b/docs/articles/04-ingesting-data_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/04-ingesting-data_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/05-data-table-syntax.html b/docs/articles/05-data-table-syntax.html index b2d9c2df..46a112aa 100644 --- a/docs/articles/05-data-table-syntax.html +++ b/docs/articles/05-data-table-syntax.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
              @@ -39,7 +39,7 @@
            • @@ -95,7 +98,7 @@
              • - +
                • @@ -109,7 +112,7 @@ -
                +

                Using data.table syntax with disk.frame

                @@ -126,25 +129,27 @@

                ZJ

                disk.frame supports data.table syntax

                -
                library(disk.frame)
+
                +library(disk.frame)
 
 # set-up disk.frame to use multiple workers
-if(interactive()) {
-  setup_disk.frame()
+if(interactive()) {
+  setup_disk.frame()
   # highly recommended, however it is pun into interactive() for CRAN because
   # change user options are not allowed on CRAN
-  options(future.globals.maxSize = Inf)
-} else {
-  setup_disk.frame(2)
-}
+  options(future.globals.maxSize = Inf)  
+} else {
+  setup_disk.frame(2)
+}
 
 
-library(nycflights13)
+library(nycflights13)
 
 # create a disk.frame
-flights.df = as.disk.frame(nycflights13::flights, outdir = file.path(tempdir(),"flights13"), overwrite = TRUE)
                
+flights.df = as.disk.frame(nycflights13::flights, outdir = file.path(tempdir(),"flights13"), overwrite = TRUE)

                In the following example, I will use the .N from the data.table package to count the unique combinations year and month within each chunk.

                -
                library(data.table)
+
                +library(data.table)
 #> 
 #> Attaching package: 'data.table'
 #> The following object is masked from 'package:purrr':
@@ -153,18 +158,18 @@ 
                
 #> The following objects are masked from 'package:dplyr':
 #> 
 #>     between, first, last
-library(disk.frame)
+library(disk.frame)
 
-flights.df = disk.frame(file.path(tempdir(),"flights13"))
+flights.df = disk.frame(file.path(tempdir(),"flights13"))
 
-names(flights.df)
+names(flights.df)
 #>  [1] "year"           "month"          "day"            "dep_time"      
 #>  [5] "sched_dep_time" "dep_delay"      "arr_time"       "sched_arr_time"
 #>  [9] "arr_delay"      "carrier"        "flight"         "tailnum"       
 #> [13] "origin"         "dest"           "air_time"       "distance"      
 #> [17] "hour"           "minute"         "time_hour"
 
-flights.df[,.N, .(year, month), keep = c("year", "month")]
+flights.df[,.N, .(year, month), keep = c("year", "month")]
 #>     year month     N
 #>  1: 2013     1 27004
 #>  2: 2013    10 28889
@@ -182,20 +187,22 @@ 
                
 #> 14: 2013     7 29425
 #> 15: 2013     8   775
 #> 16: 2013     8 28552
-#> 17: 2013     9 27574
                
+#> 17: 2013     9 27574

                All data.table syntax are supported. However, disk.frame adds the ability to load only those columns required for the analysis using the keep = option. In the above analysis, only the year and month variables are required and hence keep = c("year", "month") was used.

                Alternatively, we can use the srckeep function to achieve the same, e.g.

                -
                srckeep(flights.df, c("year", "month"))[,.N, .(year, month)]
                +
                +srckeep(flights.df, c("year", "month"))[,.N, .(year, month)]

                External variables are captured

                disk.frame sends the computation to background workers which are essentially distinct and separate R sessions. Typically, the variables that you have available in your current R session aren’t visible in the other R sessions, but disk.frame uses the future package’s variable detection abilities to figure out which variables are in use and then send them to the background workers so they have access to the variables as well. E.g.

                -
                y = 42
-some_fn <- function(x) x
+
                +y = 42 
+some_fn <- function(x) x
 
 
-flights.df[,some_fn(y)]
-#> [1] 42 42 42 42 42 42
                
+flights.df[,some_fn(y)]
+#> [1] 42 42 42 42 42 42

                In the above example, neither some_fn nor y are defined in the background workers’ environments, but disk.frame still manages to evaluate this code flights.df[,some_fn(y)].

                @@ -216,7 +223,7 @@

                -

                Site built with pkgdown 1.5.1.

                +

                Site built with pkgdown 1.6.1.

                diff --git a/docs/articles/05-data-table-syntax_files/header-attrs-2.8/header-attrs.js b/docs/articles/05-data-table-syntax_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/05-data-table-syntax_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/06-vs-dask-juliadb.html b/docs/articles/06-vs-dask-juliadb.html index 0a92819f..55ab6931 100644 --- a/docs/articles/06-vs-dask-juliadb.html +++ b/docs/articles/06-vs-dask-juliadb.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
                @@ -39,7 +39,7 @@
              • @@ -95,7 +98,7 @@
                • - +
                  • @@ -109,7 +112,7 @@ -
                  +

                  Benchmarks 1: disk.frame beats Dask! disk.frame beats JuliaDB! Anyone else wanna challenge?

                  @@ -137,29 +140,31 @@

                  Timings

                  Please note I have not tried to record the precise times over many runs, but I aim illustrate the magnitude of speed of the different packages

                  -
                  library(ggplot2)
+
                  +library(ggplot2)
 
-df = data.frame(
-  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
-  timing = c(17.9, 50, 76)
-)
+df = data.frame(
+  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
+  timing = c(17.9, 50, 76)
+)
 
-ggplot(df) +
-  geom_bar(aes(x = tool, weight = timing), stat = "count") +
-  ylab("seconds") +
-  ggtitle("Convert to desired format timing")
                  
+ggplot(df) +
+  geom_bar(aes(x = tool, weight = timing), stat = "count") + 
+  ylab("seconds") +
+  ggtitle("Convert to desired format timing")

                  Next up is timings for the simple aggregation

                  -
                  df = data.frame(
-  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
-  timing = c(1, 1.2, 6)
-)
+
                  +df = data.frame(
+  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
+  timing = c(1, 1.2, 6)
+)
 
-library(ggplot2)
-ggplot(df) +
-  geom_bar(aes(x = tool, weight = timing), stat = "count") +
-  ylab("seconds") +
-  ggtitle("Count(*) group-by timings")
                  
+library(ggplot2)
+ggplot(df) +
+  geom_bar(aes(x = tool, weight = timing), stat = "count") + 
+  ylab("seconds") +
+  ggtitle("Count(*) group-by timings")

                  @@ -168,11 +173,11 @@

                  The data can be obtained from Rapids.ai’s Fannie Mae Data distribution page. I have downloaded the 17 Years data which contains dat on 37 million loans with over 1.89 billions rows in Performance datasets.

                  To download and the data, here are some examples

                  For Linux users

                  -
                  wget http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz
-tar xzvg mortgage_2000-2016.tgz
                  +
                  wget http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz
+tar xzvg mortgage_2000-2016.tgz

                  For Mac users

                  -
                  curl http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz -o mortgage_2000-2016.tgz
-tar xzvg mortgage_2000-2016.tgz
                  +
                  curl http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz -o mortgage_2000-2016.tgz
+tar xzvg mortgage_2000-2016.tgz

                  For Windows users, just download the file and extract the tgz file.

                  @@ -183,25 +188,29 @@

                  disk.frame

                  First up is disk.frame

                  -
                  suppressPackageStartupMessages(library(disk.frame))
-system.time(setup_disk.frame()) # ~4s
+
                  +suppressPackageStartupMessages(library(disk.frame))
+system.time(setup_disk.frame()) # ~4s
 #> The number of workers available for disk.frame is 6
 #>    user  system elapsed 
-#>    0.04    0.05    4.64
                  
+#>    0.17    0.12    5.55

                  We note that there is some time needed for disk.frame to start up all the workers. Next we try to convert the largest CSV file to disk.frame format. The file to be converted is about 2.2GB in size

                  -
                  time_to_convert_disk.frame = system.time(df1 <- csv_to_disk.frame("c:/data/Performance_2004Q3.txt", header = FALSE))[3]
+
                  +time_to_convert_disk.frame = system.time(df1 <- csv_to_disk.frame("c:/data/Performance_2004Q3.txt", header = FALSE))[3]
 
-time_to_convert_disk.frame
+time_to_convert_disk.frame
 #> elapsed 
-#>   20.27
                  
+#>    41.8

                  Now that we have converted it, we want to a count by the first column. To achieve this we use a “two-stage” aggregation strategy. Note that use keep="V1" to bring only the column V1 into RAM. This avoids the reading of other unnecessary columns and should speed-up the analysis significantly

                  -
                  time_to_agg_disk.frame = system.time(summ <- df1[,.N, V1, keep = "V1"][, .(N = sum(N)), V1])
+
                  +time_to_agg_disk.frame = system.time(summ <- df1[,.N, V1, keep = "V1"][, .(N = sum(N)), V1])
 
-time_to_agg_disk.frame
+time_to_agg_disk.frame
 #>    user  system elapsed 
-#>    0.14    0.03    7.39
                  
+#>    0.07    0.11   14.30

                  We can inspect the result as well.

                  -
                  summ
+
                  +summ
 #>                   V1   N
 #>      1: 100001458647 111
 #>      2: 100004788186  95
@@ -212,51 +221,52 @@ 
                  
 #> 389486: 999981910757  17
 #> 389487: 999982397951 104
 #> 389488: 999986952752  97
-#> 389489: 999990008973 156
-#> 389490: 999994125744  15
                  
+#> 389489: 999990008973 162
+#> 389490: 999994125744  15

                  Another way to perform the analysis is to use dplyr syntax to perform group-by in one-stage which is:

                  -
                  system.time(df1 %>%
-  srckeep("V1") %>%
-  group_by(V1) %>%
-  summarise(N = n()) %>%
-  collect)
+
                  +system.time(df1 %>% 
+  srckeep("V1") %>% 
+  group_by(V1) %>% 
+  summarise(N = n()) %>% 
+  collect)
 #>    user  system elapsed 
-#>    1.82    0.25    6.61
                  
+#>    2.00    1.41    6.74

                  However, the dplyr syntax tends to be slightly slower than using data.table syntax. This may be improved as much of the overhead is due to inefficient use of NSE.

                  Dask

                  To test Dask

                  -
                  import dask.dataframe as dd
-from datetime import datetime
-
-a = dd.read_csv("c:/data/Performance_2004Q3.txt", sep="|",
-                dtype={7: 'double', 14: 'str', 15: 'str', 16:'str'}, header = None)
-a.columns = ["var" + str(i) for i in range(31)]
-a.head()
-
-startTime = datetime.now()
-a.to_parquet("c:/data/p03.parquet")
-print(datetime.now() - startTime)  # 50 seconds
                  +
                  import dask.dataframe as dd
+from datetime import datetime
+
+a = dd.read_csv("c:/data/Performance_2004Q3.txt", sep="|",
+                dtype={7: 'double', 14: 'str', 15: 'str', 16:'str'}, header = None)
+a.columns = ["var" + str(i) for i in range(31)]
+a.head()
+
+startTime = datetime.now()
+a.to_parquet("c:/data/p03.parquet")
+print(datetime.now() - startTime)  # 50 seconds

                  and we can see that converting to Parquet takes more time than csv_to_disk.frame. Now to test Dask’s ability to aggregate a simple use-case, we can also load only column into RAM to speed up the analysis.

                  -
                  startTime = datetime.now()
-a = dd.read_parquet("c:/data/p03.parquet", columns="var0")
-aa = a.value_counts().compute()
-print(datetime.now() - startTime)  # 1.2
                  +
                  startTime = datetime.now()
+a = dd.read_parquet("c:/data/p03.parquet", columns="var0")
+aa = a.value_counts().compute()
+print(datetime.now() - startTime)  # 1.2

                  The aggregation time is also longer, although in practice 1.2s. vs around 0.5s isn’t that big of a difference. We shall see how the differential changes when dealing with larger datasets in a future session.

                  Dask is good

                  A good feature of Dask is that you need not convert the data to parquet before doing the aggregation. This is a very convenient feature that is not available in disk.frame yet. For example:

                  -
                  startTime = datetime.now()
-a = dd.read_csv("c:/data/perf/Performance_2004Q3.txt", sep="|",
-                dtype={7: 'double', 14: 'str', 15: 'str', 16:'str'}, header = None)
-a.columns = ["var" + str(i) for i in range(31)]
-
-startTime = datetime.now()
-a.var0.value_counts().compute()
-print(datetime.now() - startTime)  # 50 seconds
                  +
                  startTime = datetime.now()
+a = dd.read_csv("c:/data/perf/Performance_2004Q3.txt", sep="|",
+                dtype={7: 'double', 14: 'str', 15: 'str', 16:'str'}, header = None)
+a.columns = ["var" + str(i) for i in range(31)]
+
+startTime = datetime.now()
+a.var0.value_counts().compute()
+print(datetime.now() - startTime)  # 50 seconds

                  However this strategy is not particularly fast. In practice, you are almost always better off spending some time to convert your data to a efficient format before performing analysis. This may have a time cost, but the effort will pay off very quickly.

                  @@ -264,22 +274,22 @@

                  JuliaDB

                  I really like Julia but it’s a shame that JuliaDB is no where near as mature as either disk.frame nor Dask for inputting data. I have not been able to load the CSV using JuliaDB’s native methods and have resorted to using CSV.jl to read the data and convert it to JuliaDB.jl. Also, it’s difficult to figure out how to run JuliaDB from disk, and the memory usage is enormous compared to the disk.frame’s and Dask’s.

                  -
                  using JuliaDB, CSV, OnlineStats
-path = "c:/data/perf"
-
-# read data using CSV.jl and convert to JuliaDB
-@time df = CSV.File(joinpath(path, "Performance_2004Q3.txt"), delim = '|', header = false) |> JuliaDB.table # 48
                  +
                  using JuliaDB, CSV, OnlineStats
+path = "c:/data/perf"
+
+# read data using CSV.jl and convert to JuliaDB
+@time df = CSV.File(joinpath(path, "Performance_2004Q3.txt"), delim = '|', header = false) |> JuliaDB.table # 48

                  Once the data has been loaded, I can aggregate as follows

                  -
                  using LambdaFn
-@time groupby(@λ(size(_,1)), a, :Column1) # 6
                  +
                  using LambdaFn
+@time groupby(@λ(size(_,1)), a, :Column1) # 6

                  Unfortunately, it’s much slower than both disk.frame and Dask.

                  As a side note, I can use OnlineStats.jl to do aggregation (relatively) efficiently. I think OnlineStats is one of the bright spots for JuliaDB, as it has powerful online-algorithms that allows you to combine statistics by computing those statistics in chunks and combining the results from each chunk. It’s a shame that JuliaDB is not mature enough to make it shine more brightly. For an example of what it can do, see the example to compute the mean of Column6 grouped by Column1

                  -
                  @time groupreduce(Mean(), a, :Column1, select = :Column6)
                  +
                  @time groupreduce(Mean(), a, :Column1, select = :Column6)

                  You may have noticed that we have not converted JuliaDB.jl to disk. Here we will try that

                  -
                  @time JuliaDB.save(a, "c:/data/.jldb") # 26s
                  +
                  @time JuliaDB.save(a, "c:/data/.jldb") # 26s

                  Together with the time taken to read the data in (48s), we can say that the time to read and convert to the JuliaDB format is 48s + 28s = 76s. The issue with this conversion is that we are still not able to load only the column we need for the analysis. Random access to columns are possible with disk.frame and Dask.

                  Also, it’s astounding that the output file is 17GB in size and the original CSV was only 2.5GB! Loading the data takes about 9s, but uses up a lot of RAM. I think it will take time for JuliaDB to mature.

                  -
                  @time a = JuliaDB.load("c:/data/.jldb") # 0s
                  +
                  @time a = JuliaDB.load("c:/data/.jldb") # 0s
                  @@ -306,7 +316,7 @@

                  -

                  Site built with pkgdown 1.5.1.

                  +

                  Site built with pkgdown 1.6.1.

                  diff --git a/docs/articles/06-vs-dask-juliadb_files/figure-html/unnamed-chunk-2-1.png b/docs/articles/06-vs-dask-juliadb_files/figure-html/unnamed-chunk-2-1.png index 6e8deff1f90797d9aaef7e7da5df7c776237d309..190fcde019b5c163b70d4c5901296dc9fafbc86b 100644 GIT binary patch literal 28103 zcmeIb2UOJewmm+^s3|IGY7~(q4x%6u3n1OZSOx?X1f(0KNLPA?#H%JzM299IU_(Tj zQl+Ct5fD(2Iur{}@fcft;n)~zp*ZTe5y6?Sv*IU2e%UyTbVCFlYa?U<`?{oNl zxAu+&pDg=?!C)+4?fglX!I5Cc7pytE2QhS6@&2&gZ0x7d%eyNwE3RbcP~<(Y5YFORR6=zPhFX>sX41kvtqVv z;%Q@Rx7SV)wr3XG8eRQ3uC^l7*w!m~m#UQQvNG+E5v?Dk7b)&nI9qsShhF9vc^|u+ z_}N-8^R!HDWA(9F$ zwEOc{AJFf<3JVhc&4mje77G7mj>OYB^qbYIXUWp9TYmW1YZm?H6OGbO>DS%6KcLUd z2>SJlf8ZsQ_a*JEI2fWD;T*{4@EQZB8wc|H)0&+M1~V$x6?9&^e7eFa(kwzU-Y?^F z<=+fajkBfN3OscthT2BD6-x3hSK8)&{N<{!{nxHq5E5hokSt6X#~;&g^g+%V^W7M^t*^XcF>3l=QM8>v>_c<9lEfr8;; zUY}t}w1r@@F-qCfsUgiQhBwmFQ_LT#{M*N0W=ypE+Ti6X;{{Var`N>@I@l7LF1P0_ zad7o_56jQbe^Os>)t2urW~Nb4-`E)0zb;%u%&dBG(vumBTqm=I$^$q2xS{b6izeD* z5;P;Nn=`Du$D3{R3=Iu0+Oebr`5pTLInQ~;G7TSML4W$7ec=tS*Uy)`Jor|zD@<=g z;LH@W{Iv4Wxy))M=|*#3$5lFU=abWmmBgoC?x?JM7^kYda{H+V!3yn7d#pmo3{3+C z{EnKSS+iX@j=7E-{eO^XJ$!dTl%lIO`$|rZ{EM=X8pqtW2=S?>^Ax8?>y&eDXM z_Z?n|rEwd4`9b@k#WKcWcv$81LaqkeGPlj6lD=W9i0UReIXSJCINzZ>qhderb+?}W zY8p5_#$q3NaX-Piqgb)`N@$MHAT!x8y?ROUt7DF7ve*8$M(=7^V&LnO7My>Kbg}f_)%N)E4q9`~kQJ)(`2&q5Gh?PM< z9@lmTPsMALm>16A?Y{r%uEA*RuwUZxp@mB~Txome#P%Dsx2Sye>~3g^cuC_TlPw9U z=0*7`ZJ9PT3sbNcj6<@sh7B@Yx+->^-+Xh*i`!W~xX?57%GIl@)F&T@RD=xUY1i(% zap97j@cP{Q?rEFGRqfd5$;nA(dFd~+U0+^2;p8GdZ-v+X<09g5H38$9ynC_%XWNT> zbDN9P($?@v*&w^AMttVg2^aMf%UyOBaB{a;XPJw z=vuAVV;O1YQC_V)d|oGj8yY3=Xlz{-9X`}nkci#xva;~*3g>3vHJm5Ro{D>So~;_P z*9__HKBwSpUOYYM+;idRLXCJe|BO0q<*u#ble?Yrx*`lhyiKHzw6ZN{rg(YX(e^G+ z_Uy3giS^X3iBlbk$SstRJb(Ddw33-g?&ikSsPBLN(L1o_-1^JSQ{!=>GGfUZjt6yl z(*mF1iM!A5ovf^x=&Mhjn#wI6OI;k9`X}$!8h+ku-`e97bv<#u>%NrKUOU`#aq9`| zfu@_h7vh2@2eS~>@qKCnUX9JmM;CjWGi_*Ne*aR^Y1Gtqpx)4;idJ!B_2!kngV}~Y z6GO4sE1MeCiHAvvana?Wk~a7LSX33kVK@3~%4e4Z#}13#bXG{XD#M%aQ?%{m;x69~ zzWMgxW#NOl9ct5)W0fb=r=IPK+K^BfSK~kA+U0Ke+#R`Mx4uNqAL|d^4`bth>k41p zxt>1U?mH|qHr!ECSluw=KizL3TVG$F-Bjk?6Km*{i*>KX)3sS!C=cB_Xgfa8Ts4`L zp6;w$*_!97U@SnmG;mV)=5|_-#Cvx~)!z8|u6(|~Lc7<&q>C3X){mV(a(`)8j^It1 zy2k8z3mqEDo}Cakk5+W;-riW>;IoX=S}b*Bm%_HKTetSsYRa~~Twl_kR!|W#Z^5!B z*74y=P92jCx$-SL*tK((nC>=4R6FT!dieI7E|0oI9cTHHqMq&9IX9Q_0}t{qW(P)^ zy09|4q}x{Y#GNL}T%IwrG^GufnicH$cVf)zkA>4q+lV_URJqim)}pKscu0Tp*a? zkCe_9xnh#*lse8kChPaDLhIqJYzkB5{0_m)Xm6b)K3xqR2wdp3Mj1@H$mV$Ufvh_1 zE~^5!UiPQ5h2QtY2L`am?rNULjY&;Q#vL;>_qO7_KDYj0^uxecx97S3`T348@!?x9 zPA^sVa2Ur%uQkb8D;O3`mrNn=Xs}O-Da!{=*2TL_k2NMW@OMgP}$j+!*MAS^X#Pa`sVeE(3FxpuglbDj(GEe$4YLm@Ey2W?yvvi zoYO-io7j5GeP{Mu-rjX3R9j=Xy=WsM#Z+0~)Q*U?M(i(jRBoD0>-XI<^K5gcyT%^t zX;_nZo_$nAZ8OKY#GgKFt80yFl$^Z*yOUcftAi7gcI){cRqY}yPS19^ob^dlGtO1z zlJTa+8Z)DHIyEiX_MuE1dXY-Q6GNFbYFZIV4ZNoI>GRC^nbAa7{$Q>oOIxY;O5w2o zaAmZd{RSLb?(33HwkCJTvqag@%w2zl*VA7{hdb15V%-heU!V5uOb?v4Pb(U4;p*R~ zw3WI4_OGM;jp@z#8VR~;Z*Gf3gqX4YMlXw3;Z(;3Oix<9etyrJ79JOuX`~+Un)Y*Q zEIqJls9-qGa>RVXG0813*s;o=V{XD-nsxiw67zGFvL%I6UV`a~8Z71O0qhnc+76c& z!3y1o&L+YKNp<$`w{8UdyHW4%c4un)4L=@tmElLKBM&#~n>3~vWk#p;x7er400t=7 zPrSSwBA++uH1+0nRm-MgA1RhCg3s!rH_uO2wN%Bcmuv#~kWU|v0p#$WiD@6bDxbcW zk0`az$(LI$sSuTm=^&a7fYw)oBS#pw(EhVptewjQ6qCbt0B)7aGTMe;n8)A1Nh7uvRzHNbm>8^l$_kIH(tU*k+QP!b7y*TL(x+eCVJxMQe?CP((sQdHB zWn_@M?v1IwDxIy>XPo0;lqnzMGRN7sI#?qvCcIrRSf6fDvdVB*&XwAp7gq;*BHBaE z0~c=Ev?&a!;k>n>HS$@!qlXFrntqE-ynlsb8q)9LM&IG0i3ra86)LwjIky)s=Fn%& zLOhP(oEB9`Gm4=rt`QjHl0DGSk{hE}*Bvldb@b8JE?#zab^?E>Al)biI49NZ&9f5< z95q7%mK-Hy;zyfXcj+4A)95$UJ$kge1*k-RnSsNr=O<<5llt>|;_NpF#w^k-0(e~p ze7mVG@1`3+%j7JTv)h|jrKh7>Z)ZVAuyQb9qM$=A%5&;5k9($k@ZE%&X@2gc>&${2k5ZG~gDoqN788(4BKWs}2TdFcu!^hHlbjuurQBQT z%DJc`CkxuG)v*#Dww({oCg}M0=_N#(`y@3_X?^U)adp~jCz^ayz_C9#%UQ0vOslzO z&|k;+#{M5dJyz~G?JyRifp{aE%YU##t@c37*_AsYI8!x(DV-S3R6)mdTC2xInRH1LO}XLMllI#2H(o<{A8p7j&Q>W&eYP3M)x@Vvx&2#$?VsG!YRLI@fkI02g`Aum zsebS0ze{ov-6f0!p}hc;$yMW9Pb{q3w#002KlNeM4t=jI%Zyw@;QoToUNro?%tV&5WO3TR!RE)Dj#o|Mlj>`@L8H zwn*Bb^WupWcA3+X>b&ynVlh?vRR>$50qN3toQ|0YX0!i9o9pP~9l0$zJ3?0QW+qBH ztOLVUoJ5a{>}(J@vD@y&)xC`tf&SJtaj`MZ#fr`WBQYG)p8F01t=Br`4SWQKy4f{l zTHXs*1Wvt9EtXsW5LDzdCZ#?z*;vxvcNoW*!>OpNQ5o5l$#ZviU#jZkwmoD;?9X?P zcRf8S;*^=@Sv(Tk)vJ}~&B?K}ALMA=l!*U9M}6AO-0RsfX8#FM+kUzFRFm9RuNT24 zd!-qS(uP$}+dj!bk+CYIzpw9lM%)SUfLkj54MsLV^}*upI)2Z;?COo5>Xb~);g`-+ zdt|NdKRIg4y@f3@ikT>$DC%0jIUSgX+g2e;>8G+XU@Y0fF=Igqziqgpj!G9FWOi#U z7WZa|e~L=cw?Xz=FTVFS*+Xl$V}760xi8z;BrHJ48-U9aL{y@Ky%+SXm|DjN28 zzN!-+N0A__M&{;C*UN0qzTRmIR=t|ILYJHuN}_DVK)|^*ItmK~=INfp?Onb11Gh9? zK>mx`WUEWxO(*wq+2z-p@7AlyCR_crDkZ>jm35;0%RjM~`@Oo&X%*)K9Jf4QWp8r( zxacCs8;76W0SLYDyq|Yhd}eoCK#gyEU7FSLg<{5WGr*crsF4;s|Lw^wGfi{x(JVJB zD^zkD1E$`Xy4UL);l2^71o&65t$WjL2c%yX^N=E7Wu3HP+Cx-_f!RiS?5Qru*S{S2x|uy)`z}ro6=5>u26o6A_hHTPU9} z%YoHn^t2Tvu#T4PV`-lHn=Ujr+Jtps?yeCDwFlCT9+(qS(>zHkv|-r8y=X(pBkwl9GUXQM}8_r+1BTnz6k-I&ezxVla1$Fn0Gk2 z`gd=PFV{CT3uoqyx7f#gSorCZbq@hULYV~RGvi;V38qH6dZ|q1A(<7i`kj{F$-bg$ zXy)WtuW~h&hs3oft{sSxll>4O>1<-0&-HuW;c8CUcOxW(2xq?+qFQnBs=gj{uSROr z)lq#_60{|Bx+rgh@4=q!61g{P`6cY@{=jecuf_}=auRfZIA~^OW?g+gDr%!m_v0Pn zRjA$@`p#_H*6=ys<6utb#%r(pGXn)PgT~g#ai3AhKKAo($5T5A_IvvrJa{{&3w6Xp z+r-!il-tU(B~vb;-j2aLTA~|e>so3;6vi_<$0>B3#SN8U4Pm!Z_T9Xs;ta@!gZje5 z=~3+sk(dJqD-G>LrITp)WfsKVT%Rs}{K`!Y@!aU>Xj7lwgea6A!rH0Bf5I_m7an@% zh5F3OdCL!P(N-SJs%p8~nC_})*Sa@2GX1Y@cS&=qihmTTkw90yIeEt@R zwO;hGf`Y*u6K{vE?>LK3NH(!2dhXxd>QJpXA0Qz0%U4bkU@2U@ClMCg2V|o<`0W#f zXwG{*|6S|&#+?l(J{5f!qmXVk4HDye`HR#ei~0eFE9>+UwY4&4G$)S7w>362bo&qc zT4Z+3U9fC#T=9$3%DPhF@;r;jZhx*9+hF-%Rr&})-{g_&?yYkdED;^+(KN(1D4VBbsf2mcIv-jN&{=r?LXG`exez0gH3T8ZVuUs}TWO(3#7O&O6 zYQ4VmscdIR;M76hYa6wZs8u5qz=~YA7dJO>YTsxMJiUFiF57-=$KacnJJgNi;t*We zW1gxVuSSNc0+um9d#gr`AC@Q@lGGIC{^xpsq+^_A(@blv#p_I)`leEkw91Sf1^~qU3jmNS+6fE)OQh;E)qy5tsY(q?up{B@*9SLm$?7M!^O{F~F9la`iN?rT!(=hqEPx%YBO^-Hs1A4Bh# z5h(0!?w?wknj;ggu~a=^szC||`NDUI$$J{p%%}+BiggAe`&uDr^QMObJNjzMmaN@( zCd7Z>pw^FUVog|TsZljPTEC|(>vh0Pr=)sR{PciLRZEI_krJ^YaX8Rgea@PvOY?=aYT?{jT5>BgNzW$D6f&y0mYhF>`Sa)@73~_7| zB?AA@9*e+*d%3DtmQA~F2l->mE&Gi?Me`AjhK^H%RLR{HQ7@FM#5rm@aBtcCn_p55 zPH1a6(y0-?BMz1Ia<@l6ELHJ3tPQGizPOqS(%(Z(*`lx{y+rvuI|Wu+A-CcbK!5#= zLhG|>u)CTWYM&L1)@p8aYQ0qmTv2mr+et3YwFROlA$zJtyUHA2*q;eB*7+_jHWwAv9-`rIk%&N<1^Wa!LJhwi^EpcLey#DhtHynn% z$-d+fVgpz%cRrT1L>{S%vWbg|^=wxJKUXc+Xc`J|8z$Jl$I!6RE)d)Z3bJxsuINC1 zZlEUH56s;|aEc?ZLjqNRueFeax*tY`nm8B9jr7#$u!*h%LsD)%d~*Ey+nr)jBn$ge zd-Jz+I1EzuoyA*NA`&Q$J<%if;0iFZ_YAMcl`5;Pw~PJxSM3|qx+b!L$iqVR=r&KB zu^VkPnyxIF(xhNd)UZ*Wi`7DM zHzV3s3<444J$9EL@&t?!brgUgP%yH_iVn6HgT)L(vbyzn5P{XC7r5CzJN|yz{_69J zgT@nh^T*Hg7q8v-XlPO)F*ESF(ryU+Y@q*9Mu>W1ot+kTjUL6eX^Q! zbcN-smbcC_d53*@j`lrUz(dKH(#4}B zK;$7HXIPfAK>$in{Fvbn$J5^7aMRyrnX**@U-0ULx@j?cC`?bg z9Fj-`=x*zd@>Tgdh*A4Ik?#uRVCaMQIg-rmo z4`MyNdz3q(z_29%<~M^6ac!|pv;wg4#%87 z>^?Vt=>{f33oG5cXoKC8Yo>0GxABlr!YEHd6xq6M{9Z%r{RL2Oq?0~Ee#-zXu)O{2 z93r}0K?^4UP=fc;B+P?*FCV%}s{(lmOVD@)1T+tIvaY#D?G;wlWOAqPPPxhc6gxY+ z4L{%cTOoo%9wLD2gKtk<@tqhVqWiS!coT71-l+RCAv0|&$g zS93E_iQr?dWYv}g3f7w5x^Yw*NcK>SwAZX4ZTq|7Loe~>7d)JSPIz2zf6z4aUMtpbeJE@$_SefHzYAt{%pu2*tsFjg06HtxBl?tgeUK$EAx~em)__W`d(J z0*^J%aQ2q7V}13~2uqsmwHC!HSR66a&l%h5v+Wy4Q5~`}5XDje$+5vXPau}b+bhnJ z-6D$fQ(^HZ#Kgwi^^DzaAf&o#!cki+Oq(0@o&A*AJ@NU^f~!6Qn81JkBz5$^Kw$mn zy+mKjV^V~-WXFQ-7Bl;R5wjl`P|WPdpKty8FVT+wc^~Z`QLz7uo?_qGuJFIS0RKr% z{U3dz`s~Y}z8uJRJ9}}f%^p3y7jg?q7bhhpt3;*as|8p!9|ML|8 zrH+(;bUEqvT+7l;3L`!Yf?gGJXbhy{9_TC&kCcf;qF%_mX9Z;>9Bv6o5w$n_NpMr0 zXx#`Zhwlnu211{;w6pMyxA5Jr2YC**Wam*~9$7La0qcg@_ETSU{tLCD75A5k9m^xLA!lg-YP0uxxa|` zeFGe-<>=_h?tEP)UITXv92Sw5a5i-HXeQ{Av`0{cNF!p)E&X>-5e-Anzi{)@Anu6p zs&Uxb8|Mz+Irizuqpc!VKn1*qHfnkckAzyH1TFu3$MUoA2-pm@=5_U0_`SYI&qmT| z=x4dWVtJ^+LYcX(E}9w%wGEul=oN!J`ZgK9s*aYEW^M2oXj*(?@0;gKN}2}pK-Mx@ zE)Un2ZHj3``9<_IX_1g?h*#CWbwK#tGIh3TPh-j9(ut?FFNj2CYQts{ad&bA0B4w4 z-Z3+tE0eO2zWf4LscKOnVV-=C!RQsLKAn*N9t6(x1a|8LPU?Yzk0P%NAh+ox*jLLa zP|4r2YC-cX<}Zi50j*b^UWegeue7by#KIyTprTvL*+(w`vb_z`;Zl4R`3eNn9fB(C zd8owyW&p6R$Qj20-_RLd(Q33?Gi;8L{6zHG?S6?gWch4GMCdu zS$>NH*RDss%zN>*FxuliD}=eXt|gohs32v6^z8Y<2+hyyAI=b4}@|q z6rbpOCzsezjW^`H_3+eo<%_gXK?h!b6}e#3!u=>?+YZF63{lxY$~rW;@ab`mEL=rY z4lR~7KM$qA4Jvi>1~C7-*rc4IogQCe$>VmB;p3zur7WiH7u8gsoLnMX zT@fyAGT)2XA0}}b$^B`msWKg>QYFB@??RndkTbh!eA9|M345iyRcGJWwoK&9hDVQ{ zy3S!-STqy=Bp}xMx8t)HZvLiRa4aJ|_uDaNztC*RTM8^Ln3ke@R#Xx&9p}BZ)QKh3 zeOO3DAiEP6-Rk5>qE-y#Fp^Je9KTNX?0b?_)q;vi3AKm@3Sn}nX+#Rmb~o-RX_bOK z=UYE5{RMj(5DG$0j>qFnPjtis>|2pl3ktV+z>JKuw~HZAVEUdn|Cu zD8rI9&+tPa^FfXR1voZJV+p5>xh9VxC)YieVBGk4l~(kV#}W?iv**rlO6gL9f+_zJ z-%b`K;>#vF0s~2?x_xwB56(vAJoPuX7dw~A?=KB%b*t0jK`pIOft1XP7x3Z-pdpZ# zYN6`rZo^^JagdQywB6?n=R86{B})7zhlm7x~9_xzLg@%~K*-$e790vd=Iy)lJ%{7c%o+tt{1Cit4;^~> z^beH!TFzO+QUKv^@x4L1c^huvmSo)Q=EoAHq8b1Ricig;&Vs&cyuD}^!y(4c&{jP% zLU@z@H8uW!iQD+U^``$PicYb`uht=M2hNPdXTmUiT1xmqFFd-w?A1C@9&mH)a4H&W z;JShp;r%J8c6Qe6xt56R%@AQ--~|EKC0D;z=e!PhkdSND z1KZpiyT&=r&b8ynE7gFE@v!ivx=<`-D8*j?u&^>hrV`Xk1QIt>=oGb_t}pm<)vldv zk^^+qc$QB2y;qs#fr5!ZQe^5^&!g|Xpm$xP)E8zjo34uRUQVtvnTn@?`F65Zz*Q58 z!|uEq<5Z#C!K)_oViaT)HsRFwUw~g}F|&cig{fCkX!oemfum4G6>yk43ilnGNzR4j zJwnvCC2?m=KutH20^rFfd87wG$ z7JT;4LUgV>Hc{zQYAnhcnS;OYOgW*hGzlFd!UFq-hXlaMG^niEN^1lNtdRDr`*i|>6^Z7 z^);FNUng)x$(eX^1O~Y{V%u4?Hl3h*s~((PNy-OlXaatTAf`XPL`_6}>M*;L6n zwgs>c zPId?uwG(6|PXh+VGU*MjzjtUS@Q}Dp%Qjsz8|07mM>45<;O~1wryBMS7Fz}8kU={^ zXQ<99mIq`jyD&(1C==G`OFV5mJd%x$i7Rj-RpmgBsf;WhO<-j-J1t7BNB&qC54^SD%=Ol z50}oONRx5s@g*oiE6Lg>>xxRp1}HW8{RLq1sy&NCVYjon^EasIP>*Fv49xTZ`DB4QlXmrY+R#7H zc|AEz1NwBbBZ*}<50-Vq+(=20%JRhc041s@UF-D?Q5EPX$kS2>$>?WGdnu3w4(d_- z>lbB)zWQ2W8VRNz%~p=>ovC}Jy8e((djLtp6=EJy=qxtXj)H^3?UxIs`lo*>IZ^ra z@Y$0=L07N5tl+#}xTbyOmgI`}eJQ#SV*>i1L(?ll9)IxApeX zi0D6GGq-(XF0x^BMd>&5H;(Y{%DFPFKl?@Y?o4%m_G0_n0dYkWJ<6IBnG?Qo!f2D`;(Sup#)~|AM$w}c2&2@uA(+zpP@V2AJ2DNOl?r$bWMp0xLVv$Wb=AN)5 zaR}2;JF38f7y+ZyjqN|!qH=8D`F*;0?E_Lup(sniZ~hQQ6O+LS;n6(3`T1}9E!VGK z7pgrdu%pxi17Li~s&*IN>=Ye|nA=oT!=z#j36umNvMy=KH+T9AQj+iAzwbBf+ffO_ zO&k={pR;9I@Fg0sEt6Ao?6Uo-lYkA!jdN_jFaYiA(AaP^_vX#@6z5)qsGI+BH@%_F zvm@4fq`Ql@xTiYSkmCuiLKx4pThtCeEoHLY-Nz@?rJn z7pEM+Ec7;{a_nrLq$QI(jEwfDAyE6h+$!$VF@vIqx>k$@>ru?AvbcEGwDGSYU%?Ma z_HyVA?YWB?heT9;Mh^{tabc_Xu)mGIJ`AcK&1XD1>z-v3ZdDnPI#w+4u@@0HxX`GV zR3HW?rKGIAXKO&Nw3F-dSCjO-zHIqO@pMPQio+%sT}Kvp;mpRA85MdRae@LV&!KFA z)t*S5$7nga{*#uPNw0|y8C&xD4%J8@@z@tRvp#4?lMa(bPN{=O|5m_qfd!<85;iGq zg~PS;?d{<1V6owGON)H**DihR6=PGjP*gtUs}mOY&MUgQpiJ-j^GltFi%lGEZs+ga zyO%^d09F>=q95F#gR`$#lms}gNCax7d>%7Mb^*kfhHI-BO^;6Px$(2UzJB(u75gBX zDpfr}bhi;3G)nNk)Z2nh@C9ksgQ8RGdUA4Rl{_cx`B&23uxf=CDxaKaFmX0ZO66VU zA*yR6YR7JX3_aCfGGouStdz;6#Msx@XT0ElRrxVJY1gk;phVBpe{odRauf__1RzH!GHHgCo-(`+9nAo&r*)CP{^CJHMb2!a=Jkj>NO{!$xi7^s1Fzkj}%(k61^Vhc6^oSj;(=yJ3{n4NB1 z)%fDoVwPZLdLsxb(i^>zYq&T@mNAYwU4dI^)cO)+Gzq6p|{yO^OmOLr%QWHJBPzo-%9-3BG zR!Z)DVYnbfEy69L$(`&^kIbj;FGf(t-rAsv2ooo|utd1_P$tG{W_N<#Er&ah(zZ}V z1?tf>q9#zr79*+hWrZwizUy+0=bKiRh>!^bo|$Hc%qs4EQRhyW3MxV9hBArg%P3zJ z(@EGUsY1B2tCg{39X|c3WFcg+(YMtJY+Qcfhoc6)3g`t~?)q>&N!YG^Nr~NXS!tI9 zOl#~BhN>;iYs$Y^pwQvxig>&MaSVK3;32tIEq0Yche`=DV`ja9iG;UjgRq&CVSg2; znqnc7dOxrR2AtfCj15Bc-s;)mpYRT=uMKBDCF3!>ut4Q*bQ7{hC=-SO>1}UOj8zhI zmb0E-o$+8}Q`49-%l0w2Q@i@+==)4g1u0G1+2);JEH?pt88I|9480tlJKMkA&y zz=nxqM^|NJ53;xxwZ!Cn^%j_gC9pp%jOVxZ)WTI%I5XMLLmN}voKF{LwK#*05|%cI z<`OJw>x#Ve7LaE=N%(3W%nZ%}=&R`_D+N}3Bi6)k;GlRVS+)>C){?;%;F${=J9}W` zk>`KTU%vUq`)|wQqF--)tg>fy$uy@#g0%uYApkckQ^&~x;EF|wxGrp`#aUrXEQ^kX zMJp*Str|RR98rSd;iwZTDk`YQlmI?;bDno^FiKb}JpLEfhC9r4reL;~)c=wC^Z+?l z?UN^AxRko_LS5^^K)BmnXZz^8ZAb}@FZG1q7Z%Q*TL^f>yDbuMh6IJ=EdOVpT0A(R zHhR$%&OEe+$yzO?8O}3qH;rne`@JQe1P;$Pz+UwAZ+O9 z?|qUWh=2(c07$f8Dn1RGZO}VHHh;)|#%#~KPtlwJW$xnpOQq1}rbnHlp#FQPTN4nd zL?fO&fI{Xww`5udt3m8NnS0+q1nn-BGAuUWk`3;>kz)ri-#XzTG(}DFZw?{Q9-E0w zZhU#>Erp5;x%%(N&{Eb&la2N%xjqQ^xlGDCYbxKKMroWza;Y-Q5r!MD1&*BRfM+84 z{pTlPgiTO3DLEM@5iMVSp!0RuFx$qj`TKj|#vtK_qIE4HKVR5_I|AEFc71?rPj%I` zJ(oAYGvzkasz?USHScb;rkTc(My9V7j zr$r5Mp0Fp$D?|;r_3CmLtrWg)#dcuOb_5=3*3i6FfEpyVwg=Wa4R$MR*@|%0XzB|e z2CoZHhKj5Qz2c}!eEO1-Qd6twCg@L2Q==D4GB%7!9o{1dC*gkug}v;c2}IysI2lwT znId76B#=BUB`r+}h>`%S-7G_RXoBSQ@a3qccd`B80oK~K?VcUqVDH{e_@i{#1~$C6 z6>dJF6LB!1On92^E*ozbh50h5K)D7 zgf+>IhNe*_%jwp$aP**w^fqn8WA;+1%>cjlA+y|AS4;ffp4;LQ6ABcW8ban4Gl~ z2$FP!mENO^VGh3wvoJ#=x$T2?1hsDT|7*h4(~|r)rT-{xTUr3NtQ8rSWDVmSIeB^O z)>}vDp*=u>P-_#o0|%rdgNk=|NgH>GN*yWsf~ zfp+)4OIoow7KqS_Xs&~)E1tRZ`~6$ko;Je)h`}1{DROEO1G6i7(;Aiwx(*3)y}!{# z@DA!48b;!7TcIh4d|IJQ5)@-wT3VRp22pjXW986GZ&?jM*@7W=itt%iVkOk7_{f8;QMxnrfB{s{!7eLBnCIEht z)F%QojzZ=Gr}qAnOg~9fLk|qnWo=W_IRtbfRd%xJK-@jPz`hrXWh7cMIV?6g?f=bp zfrR#R`ZJV$cZxknkM!UWF{-^GeJv`_CH zHT|oqmJ_v*D$sNx!=h6TzjK!BHU>jf62(?^i|{%BpY+%LC-i`S z!h!e~&4c-Oj^)4fDXwe!F1b+*aA;2=kU!1>RG2#>OgJdAQmkLWaC@1&{2S34w%5SG8vG`#6K{)fSqkt5M0)kegfwA!*Qb zF05*PJ28PQHhOY0B~U=39yzs4?z)xyaED-KA|hvrh2Q#()(;<`(ZTSOBhu2+L}*H~ z(n@B0GKJE4P_SnQo{7VPrYK* zQ?ppal1}WE4x23aG8_L&cquzjWm(#gTSIZX9w6^wDpY6Zk*2$y|XID zp;j21rq=z|*=BVc@$IwR7Yv5@K14edRTlmJ%72P=)d26&2nV8SCdUUk8AUiEtd_!? zLpU1jPE^|tBTe}1War}0Q*btdCCFYQ%FsxIMlxvjf!n^SiD(8R<@#mS+oUVhyYya*=_ET52K&hy9NjZ`-ylnJ+)Dq@R7^ZS|+oT>aD$+f1V==&(%lCu$ZM znw~|WDdKhtpP$L1)NLJ4MF*DGDYNMMbtyD^@V4~K30iRP>p7wZAZm?QeNLx1I5>C^ z-ssZ%NqCq|J!kAMubthFfIfu|q;fd@&y(8%y!E~=>BEZ_Eh4oJsZbYAb-Oo|ikW2I ziRo#f(gX!YPO~{m475xDkaFyt^I-WAcmQVUR1Ka`1C0}?C@RVvi@ z|FdW;J0Y0dx5VnpS2N&J=pCtbpcA^IiPqP=bdbx(;ZIH?*4Z_#dJtIWAVf@sT*mG4Bh z|G@aK{47sIDo2vNS_(d7s0AS&<#>L~c=YVtZQ-K)$3}K+r@zNkt`wXkwhIFCC$<&tH7P=-1pKs)r&ir-A>8mu3pJRG~M>H*yvuwGT6B zd?OiKmab1@+$eU-T5kpGpM6n!Q0|`~sGhXL<`P3CA&zCmZ%5_?jomw!(VZeDyj=ft zD*v;?{r=~M{^QPc3YMZrktMsyW&=3J1)bNmQk=JQq59M-Awarz?a(k+ctnlspUrLu z32+|fTf%cFN7<8U@rTr!lo18t^Kj*#!j$x0j}s$l1rwe(etcEb}3y=*Y!gV}7CoNDaYfrjc(=16xH zXTO~Aw4@|0ESfuv>njxFq2j{d%3?z;k`KA*(3}x*;g%eTMQ_N!IScha*wPE9??vZ1 zjD0nyq!}r)Vvf6P9UM#n%^Gz2;vw@hZTafw5H=)P?S3Ow@*%HDbOBM~lnEr}Tmv<> zSoks^6cURx85d4{ZOGVXq119N1H@2vd+yzJk6*PfNz~uFH>)L4@l9%addD|!AJp$N zF`v1^e50EOwz03EEyZ!17&c<4rp}ctbLL& zI@m2W#ImZKnth{z%lsY~<~2qe*bv`Ioy zo2b7yyZ!%&nT&y}T9SSgeszxiCiho{69$HcJ*PHS?}UO3eM7_wll#cvPrL!iW9Xpm zR+>|YxpMtCjrVU&zIN?_b&O*~XPd=LA+$`ww%f+rBU9joGcnJYTWah(tl5BWSut6o zqZ=Nf(KEXxz0xP`^5Mye;doZ)sxH)B!s(YTGZ`NOD7M^l?{u}j=jyf*!T`JZTQE-z(05D>c111_ilqypy#_}Pc`&) z4ss*z_bJ?gp9*K+YZR^$sHYptaAa#hk;O2V!VEzc1CO+ig1##N!8U>+RqI&Poo@v! zf2ijV#_H=Z7=`e-ktOGuiYS3cf}Q;?$4YXY@s~^M_Uv%&w+V{glwUv_Ub5OBDCHx+OyDPQGi1@$UDtas5Qx&5)sNN$@ zup8RFZR9dWMLmq|p{O!w+R9D@QTCW?n^}D05X7Zch+a`j?sg=ZkuekV9ME_-xViNH zb?@b0LjWNcrn^!DsmxvBhTV`cs<7?_wXYk?HfN^Q{6tnIGNTww48x(DgINYN(o+!{ z1WC$%7g}aue~^Zb_JCYvB=fNP)kXFX=3V}<;zyX}k~vG@_J7x|L3TnH(5Br`X?2Lb zNKHrFf>JoR=SA(G8c*YbbN~^fxaEoO&T{gD!FsPe{^5#QMl{mb^>Vi!C@q ztkc=W@t-$zMP18_XrSS^n14Vnh#?GUnSyq!jkboyTWR(F<72Ng=X_YWWhq5bXbU;v zLN*@SmN&ALOS4sE7f4hMkjbt{GP!!;42O@%vxFE_JihMoJp}MjAtMWib9;3?;KwM{JdB0dfcQ!VJd30nKAP=2x67^Cs+ciXcDKB7OgIPf z@+i1Z8f>;o`#e}({RB1iMg217Oe% zOHL&ktAsJ2;oY+t_V?zSU%17X7%Qgd^gf2`y^io5tFtPjK<>7Yc>89@N zXU;ef^=XC$`Ep+BFI~%{jYDlN<&d^YyNkh_XHXhDKtZ(6vaq**^+A8+(!wn#XmZ+G z7VWcsLq3KiBop50hA_&4Pa*>wdKH|5-ql^@bThHxX-`VP-OvmtQl}(@9ztpuhYoBb z98(EtDmj%32p$H@#UXi;1@ewrP`ww(z7ZPEi)kX_T(AS1q3b=M0Vd>NP91l4G`s6w z;OIPyFtvtUn7FZ-_`uY% zBOJ7gX%7iGIph9;PK>V3>4p$^8Q;{4*_g#^s$HvB{W4Yi;K745A_iO^6zToJ6PfMn z!s0A)XN%DgUtP#2xJ;zsK5Uqz{u;7fQWBscD-pRjB!_2?&yeGXjDPfjyXOH47I4DUyy0*{k1v}hj9x{l4H`O-*cEkvk8K~ zKRZi>kzUpGWB{{7gyP-7(~7Hri_i!0fMgeHP9&c$4K*NJlorfmWUmuU2xcs3;u##i zvNWyGj)qlaW({9faI$>+)Y-g2MCj(UQ`yE-X*2%nsDb9^3UK z0*eNQZ*$3;d$Jk8hkm`97GyeX>}%a(3o68IH6DiA&e)dNDA1Ld6@z@xi8;5%y{n{H z3sqi~l8X?OrIXRt^(=ZMW;0C%?hAx9O6YFG#wC}04py!L%+m->=fVeZ^nNq%f_n)j zE`}D0o|ZpRAZ%eg52cO_4wqk~47w|?8O$aVzVpYPp#l^2BaZj4TGM!7{L!<-fKP*| zY5v6UavUd_{7)wHZS=yVQ{f)ddG=-do+@y@<{Zfn3abnsdO8Y>g9cuQ# z09I@2r&Mx#DK#xECA9&~a7?4#+kprTnrs?A4&QIN#x!zaA9XC4c1g@30uFtgb-Uu# zR55A9i{55zJdaVKKQho5xX?N@Yi;9=RSn(wv0*bU)9p($hqL7m>SZ=qd&o_6fD|*d z?T1&HoiggmVxyf*;Lw~XhdCYE+@ZCWj0SYk@zsB8i2FTS4k&$!*E2#&G7kf0i?w^^ zHmGZEVu)2!f2U=5$8QSuhYsz;HN%{nGsu)Z7}Y__qCu}G(peAEvamUn8iY+8v-Js0 zP#xBFzu)=H*a7^Afvxm20^+BEpB$oW$H_&SsPp0OVoZdIh8f7JG-y_8fLPY0OW)Bn zl{mzu3Dt%c)7;Qnivqa7MkWR`uhJ(|({c4GYPOEaZXU1cNl8o7W|K!KwRo2!1O%2r zEbr6OeEx;T8Ien=HkW8XCT_8XEo^dn2zOjKgu37qdPtXrItZKLMQzlSDb4>9Tcv?7 zYDa2g3KUF|nql*GrAaXW%+ij-;E7SC(15w{91*ly(m2l{e863CDvC6qBml}s55_}! zW3KyoN?|l~%djO8bJ6df5UIgD-t!n>C{=^PQ^Uf0&VH+^#`gMDktVqL<&^&VoSeyH z2f4S)uF5G|+yCrj-|jmAFKbGUv+HiC2sLSD9D6~uaoff{_6L7Lo}<|{TNf%aSwgcn z-Ru#V-2-F$7!BgHWFKAJl)kSIfkae zKa-fla6Gh73jUcvd6JdffLNawKlW(AEN4jLKe=3Wgd2Xkp67CHk8e_HVJuyH|g zv7yNiPVlH2$Ge)PP1|3`*2&#OxL?JB^Ytv~d%2l}Z+9Ed=zORY1}o3^EE>TRF|Qus zhz2FH*fgkA_yYz&p_4%qABt7WT~wdrbS1j+t`g08fxErypg13`syM4=Ngd>?hoWsm zb70^J(*^V(7(5-gJ7~imef@AiV;YBF@BTrkr^QDjKcx1$E8E7~PfUIt)1Z$&nmP|n5M7au5%9+K85h&0Q<0mJCg$cR zEDH=0(JuYe?Q7@4*V~6+8eo>&z&0B(h^28fGC6{AHmO66Q?*guG3SU&>405ggeIeg z&mhK9M@y(>wne-Wgt$8FG9hpo1v^GcAI+(+^?5T1o`A;E5Br`PJJO~t-qdo)b!>0*=Od1Z3tI%4Th;{6;zgDHN zwhDJkkn0(E(I`S7UfQ8|zVfGKbyg9}&IbOmDe25iRK+f1w*$W{U9{-uGfDH<^NfBz zHmb9Bcje^+Qr}5jJ9+xYA0jtiN&0D?M)Xp*3{NqDTxuB!9E!m&@I&uWto9j>bp1Ld?2~6-U;0ijP-tY%+RgnfKeo^NC2&XSr>45b#>Ps^C(zwZeL7S)pjcDDnV&A}2d6-0 zbL<)fP*q-UV!V(jKxIfjbOK^!IBMd7*1XK3`{YYUWC4eTWnCE`IO9$86I%hnyr2Fu zj|Q^Gc@|RB4h>8I56}Y(a}|2{^$ZNgJ4!wZ(tJoit_0A;5>XjgN0Y|hLwKjDTrg{nH zP|0ZT-rvxLpTJUqgSHXyvlSfxgR+a1>k!#N!k+U(>wI&zy&?Me$SrpT+A9?7A@06U z)=zkjMP1A6N9hy`v6g)Qpsg^W{E7Ndo5@T5H;12Tv`lpoC-G=Ojw2@meNM{Q8T$@H z@zra6h280oWZ=PJPi3>ee!1eU#JV8%sLSR-k}=)V8&? zMZZG8%@ZF+u6;{<%o&40xf?&Bgq)Uy`QZpP9r*UPMCCh+=O1CJZ#&@s_)E*$`K}lD z-}#FRJ0Wvk&C1$@Yuk)ZaBl^=FPHp6qj{>I0RdD2m%H6O@i?iwyL*#@LSky_-~o|2 zjG&WSNUM!~1FoAuo*&$FOMF)|7_UNz6npeL{18{Ov&RmIKmk|!t5II}o#o{_`ir2A zs7>gd(R{R%Jf*bUwn3t&`KxqlVVj^LWmz>}>Z#0_e1KAZSB@P++Hfh!@BN4tp4@tUygp z_+bXsJnv)IW-+$ZFoh4aD^^Kvp2&)a2ri*avXCuYp>hrr`>ZkKVQ?BP*RcdMQd3h2 z|I||pmf~XM$a+oeA2ROOu zck^&QZ9tvdIyzR75v?jlAu%N-Wh*iEbjA7OgYjn z0G@v6%bXI)q(NT!`>9vl(NT+?Vo@@avR|Hs^O^y0$Yj~uqoc&`6Z6@Opg1ksHM1KO zm&<9;iFA5lMqL2;yoKgf7>nnm(|W9;^;n0wayo4MES=Yd{*qeJVNBFutG)1^KJs)d zr!;BZ?d&wD*YRcHbY*2_C{sp8hR0v@HLWUxcKc1FhK5GVMCCM1I^hVKZXRUOa1Vse zEBobOI2Jz7_ri7++~68cE=I)Ol{<#PbnhNw>aQ9@M^=f3`!dTBRy`W>Q2Jy|_u$U=0FtUC>$tkQ;<7&um_oP8SMPn# zdU!pUfrDpFz;l2U>BQvHoouv2(Et+GjPWnZuzdTW+EhIJbPF+hDJHq=z}_6P@Frs_ zO?VXRTXKn>!Gsku*L1E%m-Jz%JmZDOewfJHPmZrCmu zz>j`(M!ALWsP?(uV?DTFn~Zk@-gX+j6Y#IRDWz4y4)I2;2rEpSF0cr7TUwsaRniJ!)?~onxFA8OjLV2Ui%~9sa zJqhJqv}MZ{q-m;Q>hd->PcrxRHupyEMdnFJVL47+fsT#|j)8%Wfyh0_F`@d{v18s} z5P6%U5EyT7)D+Ovm4y6wCn5JD?N#4kP3iTA?q$1=esR+5Vl8WFY7Mv64n3E7V z1)5WxQ{A1@4HtQ%ioDH>l91y4_r68Q#VTSU1ulZ&>IBE?guvNW1)9$YKG6b70Tsgnzxn8VU?as^eJJ-ByPp@^`rG1B!Wwt@_?B!1ld5;~>R=-j0 zD0{9;2=J8&>!`u?pVRN%iU?r`_eS&q^?eJTo$>wK7G9~dy({ib@^aLP9D_{S927r> zKlkMw4Nug^&2$x%n>~Kci5f*T>fCZ@-lYwz0n{_EC{0w+&HtiPjXv0=PJ3>8Z(n*C zRNP9t-TLZGP{^lP>&TFx{_I|s?9-lLrlc>fJ)t%UE2i&NIKQd1f*^@wH1^;MB7a9f zJ>wjC7F9(Rl#<|IUAbw5EBv{0H^+)(;NSWh{U^#B1*wCDA4kfA1GWPWag?558l{?J z*<3364x0B}wMVz82g^Zpt<}?nESK+u?h>D zcEgnNAHZGQK^`g` ziNm0e4P&HI5&JXJZ+iuyB-{^@xt7aPxrzGQn0GzGSGxC|Z%C67`7IJ+eS|QX8|+L{ z#>{^zz1CEJpVZ)AMf;fQ9xk&9p}lq^Ug{vQ-?|Q2$oNJ=TDTxNQpvUakT3x~l7By_2SN*Xt)5AG+wuC) z+3YwlxYcRp=UAv9U5as5;!V1`!zE#J;lm=`eA4e%n{ex8aSbZG-_92dW)wk1%<>^g zZwn0>ux|B4au0^&KS5rfubw)>-eL*(fZk}p$hMzYSorlVQ!9wVp3}7#K#7ZVzmVVW zhj$j|jcFVf9&^Z6Poc~=Rd4W2gUzI^Xu40KdaGVSEAY_Bk()rU3&HmUH22}i{*9E)|?>aO+UW0R&7AlUj`xCYw;><$c`(yPCb7>|^utPL1I1#oi zwu+HG@wv`?D@o8ce1}zUKBOz@>-Da1y;+`Gxdt0xK*^^LXNG5|sdt%=mdB1dv(B@K zD*3!q;XQs;-bDD?5UgnH;r$+dWblk4&-Z15NenJi zoWka^$66ucSh8xd?(W54Vw=iMw}}cunI2u`W>Lo=UOfZurChC`nE5Q%8(*r16@IwZ zV{s(UZ><*CRD)MSKQk(td{=1T=*o;l@#+M5(Kx}-)5lF))cp;O?W08-W%aL?XKR^w zE)cbn{4q^d?#qFdfFL`5d-Jmi*Xw2uB9B$>fOG{lgw$BKNNADBrgJ;Lug(an2Li(Ig_c;=LNL}d888!I7>6&J{WpT>ZT*XjwsQ$7Iz z@LW5>#9*ga!6ZDl@=f+atRhydr6BFo*k=qE5rT$-jr7X|4(uioS$G>7=iOmDNiy&A zx`$bg;i-uOOyA5TBQ_MnN}vsIcV4bHd9B#0xxspGmaDy(}WsnN`{Z>DTX&8D{`L-e2YK#bdMSj@pg>g#zM+4;{h;L-$c{fs)@96#1V=& z)*)}}GrFG|F5{@`o0BefSSEg|??i)B5$9&ahw_Ev6zSgbBjq0?u@jiW{9EG^R*k=6 zO9cg*vd#ut!^s|&w(Pqbm3#9|*)7Atb)|-XKpw!N@Pe0Dx8#*7@DEK?#H}NT{ki2* z_9|dmo@g1qscTg0PxC|tOR9Rf2s~$*cTCS#qx-_zC{yi3|2tJK--YQYL$z;ayZ_H|rrG*7>3Z%a z)f;^8t_!upe)@a*mxZ0uTJn|6$kWD6!>_llrKiSMZL7&4uFX8%ecjjPt3X`EE_S?Y!^mDxekbUI4Vk9x%60zdPUVtB?!T zGcVq>oE>I3swom$%pxNtfi&c?hy9-A6npkzN0TvzjmXW-Jh$U?Q4G?E4pPa!aO0qIXzvL-& zduBG%%Qx2V2L01>m*%m1%)!Y-0r}80qJVmiu?$vCFKy|Yks#f+0@OvRnycO1n;WR- z74n-K_%lX&s|U7HrDK~xe@<(M17Lpa(+to5=i&Q_ix%(&c>kG7&MzY%Pn|zmh%5>UgqZy6|Nl`=&ug=TQ_`Nt|H@m zA8a?KC1P|#@-BD0SiO(6TMx2Zm$gGxnfItL5ks~#T@4g&`8H0-Wj@GG1?=%TK60X5 z@S53LYOI4mT32qGuLWZ@u&Z+GlN}IQcVD`{vCr8XHHTGylVg~olrs-VW_nj)9hW8& zl(4zeW2*3mp7nEsIp~kZRFf=96wdW-QL-gH?$ExgTFb!JtNJ7uh1LEe28aP8hV(D37U8*x`zq7t@J-rs5sfV|uvT}n%~$w{-rYsl-PxA}T7 z#LUGw8KjiUa5UMdbNR;7ky7FrUa#5yGSk`*2FUyp7)^xosm* ztI$#RghaZ+=Ps0kDO6ab6323S-*dOHbw-u9N8xaCo@(iU+4vHYG*BFyj2p{tVqzDA z9$w${VBLPabD)zZL<~CJhVLiRDUapvb+y$i@d9V+x1eS?FIv>yJF*}y7{F!aUrfBjUI9=Ta>IGX*F%Iw z6AYfj;HMZR=V7L#X8Cj=E(R6YZ$PmWF0 zZO?!YO}Y5>S9khm;t~yi>NELd7u52?@S+lS!^}pnmCSa(x<2F+$>^2Mk&$g0S^5vU zFFc(yWr}No_bLzb%bGmv&-bqOq*G4I`GDBe5P!(^EWU(Ybcf$UpFw(^{&3{rO+I+f zm^xzZx^2i>E|Do4?aH50!|}p8rZ^?J5>jRIf^>Ox)7)7L2!w_vGBLNKbY6OCkx)M3 z8Z{oHw!a2AQsE?5ZG=FDD0Y@rXEP&o+Fn0EWUFu=u!HHU&-)%EpZvzHb%*g8H~!)n z)ObF$vd#D0%jh=JgBEf+Z7T@X%V~YdD3p%-)LsCa(Ecd7H@FN1l!@PI(#D$6H;vIV z4OMF?C?+KPumj+cPqmD&{6qG5T)G_YBxH5hQae3&>^8G^4M3@rbL;`%KHn*n#A|IP zUB1FU^7=0istB?;;rF12+(+HuubOtEQ{qnpiXyts(tnS+FSli_C7zL`z1L<8O{JNZ zN?2pg`ow!RjDcwcNAa4@?v$XOx8vD)f2M}cd`mL%8Pn?Y_&buqHGa!0IZ+M#x&#}^ zfJB3*$<`CbZKG3ciKI~puZ=W19F0@LHcw+g zKHhqYu_~r;J3o_IH@=MSHcu`Z#2=Bc&+*(geXtGjcp=lOupU^DTmvmsa9*A_tqMCJ zqQq|`96;u6Gu+fdG+L!UM$VqgP>)8ww%i%bkH&>ISmYC4ZU>)Evveu=n9&`b$g&}h z#WBu&7(FwTr7PDgzoYm*^ugzMhZIeNaG*Wrjyz$P=IqA+MG2sYCzhxoT3P zC2pQep9f8kAItoUmU{dMROv4UC#PxA0o7@n??rHNiLfMfb^W$yf0dk-%uX0r7>37? z-SvbcN_5JWnY=|dh)iI3L@urhen_~td1eenn7j2}L=NFrjQSXUuH6igsPJBO4-7Y7 zPz!r|enzu0Pmo$0*L@M{u?w1^RY(|RaOgEPJEwDjnfWCW0`zFDlAXCm-1j9lS|1%P zrrpN_zH#KkB1JUZd%S0wY89T8%81K&0TKCLvfrSxX;v%+Cu~7R_ggR%9eGj)yH;IH z;d7Pc>Vm`oaV(X81?VcZ0Q6=;s9?q6KpZLhR8W86rVhlNO-t}4vLqrIAH>$fJV8bx(PP(N-fJu1J=1Vz zdb5A!`hPHm(i4>VO!7)B^#oIysUmthOP#HM&DLl6Fw2MkEgd)zC30qGIOS6#34%8a z3vrV#{wklVCwJsZbTRlMgwt26?Hu?53yfGp9sA`zdkCVMQpYM@nQnRyQJJ1jEh@Z$d0r9;G6S~xnbCDf0hw+iGa4QTFxK+b(}uO4B|=?a6a+~ zyk`;UM(tZ5d{;=8=XS`NxrLPJWrlpKxX1MS=sSZY9M6P5-lw$?U*{xq%x16?Ott@# zwac>qF>8OT13sLYBFxMr{s-$fA1ZdbXl#AzH3zbrtGlS6?q%~PSViA`^YJn8<}5{ znXIZF3&yv-qQ7nca+t}#d43)Xb1$%rSykQT@^#l@;zPw@v)eDA+KY!P0*AltKK)S* zTM}u?2|?CZFl22pRlfJ~dIcdUcrs`bNmd+@@AO|0y+#s}w}OE)GvLgD0tenZ zigWb+{2ihP3#0i4uS36o?SlM6PhG^?y~tMMH6pd0=L~oJ zGaZpLQQH6QnJ8YgRp};=@fw%r)0Fh%+{J7qO@(RggRouAuCX6SbQvKa{;W90dP*gqM)099Vt+o1EPOH{#1z?B zZ%~uOikXp~y%xC**uC$o7SXGz^m=aIRj-g>ot)|@mM7>CCIEP)_cbPU) zFt%8ax&^*ajJu(sS!`w&xIq?Y3SVDs-rPWenQD)`@D+yjX3dv&k&N1X^$AoaIn7U| zQ+JrUdGNqS^sCao)#v9Fw~5Et8!8AOm8b>M?vp-9o^ccVTSgXevdiCtr&t5rqPur0 z!ypbD_FwQm%K?bew36mF1f~$0{A+EqoKw4ST-sdsIJK5jYcop0`3P~%VGpp3?NB@yF1)%n zO4LpD#ed$5f3L~AH<3XJXpv=4GUCCo=!wWwTC(saV?DP&IdvuATv}|4*VO-@-z14g zL}63y5%F3QZ_#o6w9`Vjw8h|O8OGw8Q@$U5A{2h3h&J~pWC`lwK3YWn#OP3j8&x8L zu)M7{jG$$Pi_HnyqH7|6Ag=R0VsUbx$GkG_bJm^R-b{W|a(qciaJ^uR3kTB);qeBL zU41xSpU4|ger>@ay@34*Lm*_{_&2&0X|kqJ$m9IRAFFz?5~vy6HE7vb+y(H*c(5wO zaoQG9R=?fzTq9eVv!g>&HQ9!XAO{~+ZSb82Ak^q0xgJd%5YIchaVxWWTge23w=Ax{L>Kb6310} z*0RA?eCi* zQXtmK^V{Xi(ZTlpL9kH9T4ojU^W2LQRl>)mV^5AsT%P{~2&A;|MH-%#j##4?DdkbR zA^BnLKSXK;iuU7vP^JB`+iUb`Om;gQi!MW&U(qtszVPmJylivFcmFkC>*!r@)wk(AVmA8$)XxKa~pQOtR6OF|j zRPDHH5`~k!%qh(Ud7g>|_H>Eta)APBP z*Q`QI>whzJ8JSYXC!&EfaSL7w8H+qeMm+bt5a0Uj%uaQpjZ|>Q12P0O{LXNk*6>mq zW^Zp3HXr#^^@cz=jeoo;KoKiCu4-fSJWttDBC~wN=A?_nPQqWV5d=MFq+$|h#qk)C zg$>05Ia1+bpZh*wktgccsgt0`YU@D*0QDWYJ?)63MuV>Mb3?K=r*hqItNxkOp{oIY z?}_0d!%W%&GY9DEsfH&xcZNO)MArcntSWj!M4|;fFpjt4rI4sr%NB1Dq*z_Jw`hl@Wb2$30Yy)7?&j`t0_>4iC&u9Q$Wlof4;U^7r=zL^&v#L1q4&-Ui zVb!>r5TAm_jLxSqQ{*BW4g6k(03VklozEJ7w33;HXSFp~io1%H&^J~CPb&8d!xz@> zcK6JgGoXlmDtHB20*(hY^X}DP%=1YpbLvF+ NZab%6@^+rM@^Al<+_V4y diff --git a/docs/articles/06-vs-dask-juliadb_files/figure-html/unnamed-chunk-3-1.png b/docs/articles/06-vs-dask-juliadb_files/figure-html/unnamed-chunk-3-1.png index bcc2f13227c32e721038a68a6a46250494e0e1ca..65658907e03e54fced6b421c1704073b89501e93 100644 GIT binary patch literal 27085 zcmeIbcU+X`)-F7@7&Xc^u?v#KQ4|Cs2ue2^HG={MDbh6xNRRXm(XA#nbP!OGVgaNp zRbVK_SdgYvVQ2~iQU|HhziW+2-kk4`^S=9h=bZ1H-*5ZN-Xbu=^W4vUuXU~Ky4K>? zgPQvmFI=;b!C)+AegB;{gE7yY!I*RO(@*djli8J~`0JA&_wV}-f8)>C?p#0owcw}k z4V)MZk#qFFSpmHF<_yM{4Aytw>bT$RY4p0T>yoF|KWeoO@B~5?t{C0u+Ec#%fdZFN34<4LN_sH-+^?Cnn`e4~L|CRK2 zRMfxl7EKGqgVeM6`4f(>_N?SJ&rCI^b;g9nSe8fB9S_t<#TVLtTO?hwTS%p<+T2xQ z(TXk4KUr&g<=fMrn0Ho0OJ?T2f4cUNzJAEk^?TFno2%ju(LYlBrn-Zqg507iHI*b? zTwLN35{mOio4D0Horb%tz!!Xea&BK&mm`}aYlBPo|N4u5y`9I79ZSv|ssBtwahp=l z{oQ4X-H8pZTsx2VPgbV=_~blia-cYSxW-7!_QUF!HdF07Dz zvD)(HtM-v-?ah(nRj~u(-6r{{3Bep+{O zuEpes_a0-skU-{~`AbUPKNF3NI}~d%KGZq(NXYlf=_QKyIOVw4u&u|Qww;$qeixtG z{DXS5w(p0>q89k6V?1fUy7sAL&cKV32oNVva%L=&nef}=f=QmV2QNt95trI=H&F8h{gj)IFE)?$I{1Y>xv;Y1_b;@z@ckN2 zKI7FEOuOg@&0K!p9_G+qbx5o9MUVuSQ~ukTMT1YIeYdh4t9A9Gb8^PwldCy=2v6(S>N*jX_rHH}L)FKV z-NxQ8Ww5?MgFVXWP|4;68#&HQjat^4xRv7S0+~X}{X4t5yPe+tDrWKM)Vv$ptSZ*L zv1j*i2FvQK6O!0Wv2bT?vHZ38*^GBPW;x8?Vv@5hP)tiIMJrM@uN2>^qpQmkQyXJ> zOyzVZ7f+52N*Eq<8!2`0m1R}LYDM44`wM2r>r%~ahE9$)y2aw1rY1(E3*;S|IF1GMrrn=?J=jrRo>6NW##EU8&|0fF zo*DP_>C>KqnQ1z{PW4u?(fb0{DZKqI{=I1k(rrdcq@I~Y!Beh)YiDXXFg9lqvI0NdMs$*@o zsYM&>PTaXPFjnrTDr-@4xHWBm&5W|TxR;ciNQdU$>9fJ2f5%-4-`lZ{pUNvDZvPZKk@}N=y0ucN?`LZuVw1 z)1xZ=W$AjElW%{Y#6vlJYz9YF+H86tME1wue;Mj)RJyY-pscMZP!=n?qT4}BLuAU{ zs%YHQDe;0OyyE?8A07!ACfOm}M0@e`vJ6DiEJ_VKysT|)Z8NnhTT5=pT82k9hU5=L zYqKk`3hIdD(Z$jQoNITh67*9loXSgPR{5xqY|zPt7MvCEqs5PUk+*E(?CubJw>KC%(c{k-FcIGsADxK59B=gx|X zc`J=yTH8f__+6~#gL2KY5X4-^W^C4G_2v&xw;QTVe|1hcwWgJyUa85G_ZurVtMO=9 zEo*U&OIxfrK3k=>6iMp`-`+AMzVV%uEYoK=%k~+~4nEyRfCf78DJjw-z8~HsW#(ah zYYqLpY`UJi71!&$t(OQ=?^fQYKcm!oQC1xxv^$Y7L5;rMi4g6$Yhe!g*_?c?eXdK_ z`CTOund~UJsUWL@2w9aAsyJMA z$O)%J%s{SdcxI|#Cd7AYoQKOSk;chOG|rZxjEit7gY9Aa@Xc9dfy{dGT4N^-Q)C_Q z&i!WUsn+#pBBzI9dHKzX1$~ql1_nmB4$amk0jlFCVl=};nPb%{zCrc%^#^Y5*t*ip z`(%Xs+h2vfdLB?@H0*8tW&P#n%`?-Z+vMfj5d@4z>MSGWeZ_5PHw}jKy<<7~gJnt` z^lgPd&v8N^=s?D$xD_~BFf|0n3s*ZJDJy%1_d=+y0a%?UucHQGlk9Sidtyy)_ z_@9ICF3Of|QtP*+82sj(XoP8=Cl6_`Z1MJnE6i*_p;Y_`%d(paopouJdjiyaCqE46 z2v78+X)RHF7gbeVZT|G)>TJAaAhTes)M4P|ZXsTi7hl7~#KfbS6}XWXb$W?9+t{-+ zx(=JrNTueYM$zIQ<@Z< zTa6sWqf((8_6KiU8$JB)PJnCu1B>#A#-786Vl^d`jI$jtxjMC%Sjo!_?l&8+SYxN? zG}nP$p=V^L>p9*TTYF-3pjbNO>=F?}dp8`QOi#|O-RCk7*8)ux_ddO}wzkXbaPOn} zqFXd>?s!mVL#MuTs8Y+p%&QW7y?vdAEj>`?4cLLqjWoFmm>W)tY)G zrPKdy{rrB$QjVlgjE3%jvUrf911>;2V4aG3RTv(oV#8e%i#0o}QWr znSLMOm-wo$NGx)T9l|Zo)iTn@HLbTeSW-e4kTpx%l=JY(F?p2$H!RYU+NR*{7?nHb z`ex6NMjUEl9z1y9JWwQ-a?+IE-o|)0z3|c+x%dVPEb{WWI4pUzHZM6ZMy(Y;F6F&K zl025l!YAET3MX^thU=%Z!dFO}HrVe`cC*lfL$w zBRlQ$hHE%8Q+x-fW(v*$n`}E@?@dNl(pgH<2<(W{bsy8rhutSPt98-?mLO| zU>O_kYS=m~nfdzbnL)LgL#74&zbzVkA(3*e)@!&cNZR|&BHnPAeNNnyW0NCU1(MQ{ zE;R<-Mvi+LaYr`Zv2s2wE!M}LT`t8gu6=kGz%K|$B&gN_;N*N&)hm9mx~UgM3ydv(qKK+ZHQSI)0MQdbA4+M(*TEzDMWfld?uu}d5yJe%xNLj9goj0wmCWR`oXc{VoHwZ0xE~TGBR?=b2&F* z6%Zqo>WSK;SZKCQz`k`JOf8jR9rp36dq31U6ilI}K!P+r`DD3W-vrK~cS6UJt7Uhl zM!3?s{Cuwq+;c*r5r?!?GC0$@%~Q#@pQ!<7qykE0JoHH`m6y=^^~zmkyWgz|(0OEc zIx`$HJjX9$uY30#6YV~cz1;2jSLf2*MTCHNe8z`1r$hpFOKQ$}H5Bl5x96wLIf;C? z{Tov9mP_be3z-=RDM%dt>TJT)yM1aesQ5@o&JpK(lzSQ4&d25p`m)6?HmhvHz|5?9 z^RuyTvl&BERRJb?JCGDw#v4m8n~FcwWf&HKjge%}uT3#a8?L_7Qg4Gr z87p6|GL46wRVLs1EI?X%2B*+A(+;c1bD}s|x=5OH|LJsq+O&Iw_dpS|Xo1+xgYP~5 zS^F$!?-K_ADIPE8QJ2whFE@+Rux(lIFT+#zacvn{y_wwL5X+M5SA9n6#N3)ZEQ6&? za>{%OcgQYq=f-vZQg-*)&t^uG*yRo$W4lGi4~C6cd9EECELUr`dgEmC%p@%9?D?ci z-ZrIOSMRX~@5iOt;lc2DdoHZV@S4Fx%;hLslw5boOi4;AefR1Pw`w$J-BZ20gZCu) z*_{*Ns4LR!jy=7|LZnZa303{2C#haWd=kZgnTN;l<8-}*`1lt9P{rK^fU~u)XW|o* z#ib*E`fWkCauY7_IXA;}bPNb$IX==G zCgRqxwRxgxbN~lF@ni@+RA>2F4obs4b+RagBvVq>UDCHIR~|gzk)3qZBE>W>G)dYh zGx4@u#_+>%^EbbpahTW|*nYHzy+|8+|Z922k^x;$Q-;3gxa$_TXJ&`j5 z@l!jHUdq%EnMYQBhFcQ(vZdEw%ctuQkEhi4@Xq5YW3S$f8{Zt=1-2kZ&c37X`KH>e zzRC*W8@;`jCW@XnN31<>W=R}T4-GKxdG2NB8RoffpLEu7lVN9`M^#g67SGbKvpl83 z#{me1XEm?AG%V!YI?sC?yP8A;9(PPkBXlU!WV>Cp!Qihc0mD-l#eI5pQ9z;GD&8-h zcTIHW!?U`h=R|5w95OSDB#1QDGGFaFl`gk#9nA|M$d$CT-?b3H!PGP#m_kul*x0dX z1J!yDYqfdb`?Vi~w zIzCjnBdl+um{f?mjhkLxsioG~*P5td8L^LLULKuUVqWh&cev%z>98}u38@Sn1gnrR zV;E+cCAmJx3@bI@SQ)1jsyH=NSzN3Ud;B2|=8NKU!j8`s1D=d`C)zl)UfN!^vf#rL zz-6(Nyb(hJ6bP$cmo{&pB5;w^!zdFuPzpP}7Pu=N$!c_!XjbGLd3375qa(bxuQe_< z!rFBapWeZOi{$(aKhQ(RsB+u&RVdA^ZET9WHBR5(-t3)L?OWs8J*M<+<(QTk;9yqo zLqgauitX}BbK``3`%m+D^F_UX3h!OJcC8Z?UlTiQ4s{N0>f``h3l~d{b*RlCZ9U(o zTG5`*eT&lD$TAEyXCg=0zTsg;=2-tZwVSFw<4l%R!Q=qJQDo!d;aaoEeU;B8~ajAZOYly{mmI+_m-b|y#zL6FDuhILKTe~cKu2d=N~+#p05um zLA}A{_{HhG`2E)}io2D1(<|59nH6?c9~g2>ai99eBDq@QL#C=9T26=A*{s==H{O+y zG$QNP*Wj`)L4wk#wDk6HqfFT#^=OMhcUR_hF7m~TVh|!cMQ(YxQdV6Oulr6##}5@A z9Ti=SbUA%99~a~R@2sHyXKK?Y!|g(+Ck@w}F)}g=OY!DOu=M7&2mn`5ahIKQJ~ARO zklkg4_UpEIL^cQFDysl?WRN8W#@{}yR`FsMu3$O{^w~xOIkTRwYT9e>q7WW+bxln>qKP`tBAc1j;JBZt8q^AW$)?Lne$$iQ?XrU>lb3}fHuZ9&4gLgjNS`j4_sWQl4B6&GuoKNQFiI7ngqFK zbSDx;B;IH2K-&D>c_Mg&7sV)-tl!`I4kekL#~2E*-k}_J8^}XxX}^KM$ex3u)z#Id zr>gNr+x=!HnrlyFf$|h~srrGkb^Wc@bmty*gn^Fm{Sx}gB~rOPY}&$?>Zo|qmZmkx67ctUXe{or>GKf1ORPTNxgc>#y7su+;NNJ``ZFBPLj}>xeP{2W2|eV z#)%@JsezW$LKY1VPkKxZ*A60-w>MAsHp|$)zJF0vO%?D+6U@LMXuoz8RQB@L+gYeZ zE4p?^j?~+v+(hEe23ibET(}Czfg_B7LUd6Yh|I}8w*s3Lu6?MROYM)D5n5Qi-QHNC z;RBufsnKS&`Rm+|c#ibkK;7StvfErQ@$gl`UMSZ0?m7Lm1~i!Y17-aqItGvY9Ypu`4n-*)Fvk;PKETQ01WE<^52JMrqa`nAZrse-eu4eEIS z$kVb?V|(e;9gVL4ECJLM7gkYAF$xbhx6?6?3;SUU3wRbsw#JcyP4*r2;FEf zJU{Z)oqdAJ7|%%?*=vGdpoYLL{X%W&Mht z^+OIluOA3lmeQ#iJP3*5&PvPhsR+-mxM-lwbi)qpz6xtmy41K^w{E3_vI=2e`(*K& z^qTzqfuNLYqc`MiY80DyR~`d}m-t{R)F#=$MnzCY*>_dN@$CMfLWVO+<+<0`*Q~9r z)r&)<%~JB$tX;dmBht?|A#w&t^N{D8Gb=;$MzDM2yJI`EUBQ)RjeQ7G3{a1L5x!re z{0>5ZQ1oLVHBPmvm{x?bVYZ`rSo2_xH>ehS*JKf3(4=)Uz~A;^0W-+5+qiBB^~)2y zGF(iyTi!gmcPs-t&*{<>*GVKmSRu8+_0{b?l9^uE&c(fS`k#)d*XW2RP2@)F#5ZKr zzJ7gpye2EZI)*<6B%f4Wz5wed{>10b@^6k5`WqwD8Kj^1A-Q0BoCpni>+w;3B+mb5 zYqo6?1vqz_9_z3KrpC6_L;=HXpH`dd)C$>Wa2W@Nc;rJSCgDHNnQsy4Gg3sXGeE;; zVnB)y?M>I|0GStrEi1WC#6^d~I?9EM!U{4}gpqnc7qC3{3`b>&+GKyp^{wk^qg598 z%?xyFi2fAJ1oSz9JahV@;$uQLRMW5=K{dQ@HJ3I|70fzKICh0+lNqKqQ# z^aMY0t^LEVD42ks4YZiZ7?HILP!C5uy15?}V;M4HAP&lUK>Tu0cNIAMJvhAC?`~fp zf@t!Y-{ij7#=a&{!Nxwr)i+eUT#2*?nGp*Lmd>`+!5`0j&5AlZ~uJIzht=shBxj;pyXwDF%4|9;*{A;b8YM_8h zFo(>9UFG1)IFiqL7n)6V={pr|2w}1y8wHuT)6x#wwsXe8vV%zEG9-8b27NJ}N2B4NpcAe!!RCMoX|FR|eirzU(D!Gwplh2R(b;VJ39_>T%kw4sp6)^<~=fh51Q|(b>V3{O$0`lPv zMHg#doVkrG)G89k#eV^}ffTsWz$I*k2>7%zMG2M+kee-t1`~s(+~7a%#HnBVFqJiq zAdW1dCzUp2I~@W7zv|0>*m{{r*UshwKZ0#T-L(#R6Q~fy(yH~0qQO|PWSrlhKW-rj~f%b2gKkNhW@)gVKK6$vH-?+Vl(BHpEeXOau+aI&LQ zNJp4b>XXY;n$;$=o0zLXsSwlClOZS>K;ks-S^gkYaL;`u%N8%wLnUpGL=5(jckj;S z7pdmOHM!-)m3n~Vrz5Gi4H^+Ch=9Ju7gx(VmTY}S%nnOKLj&<76(Fk}`dtZ-{2<7) zMb{U;Lk(|zcJbO^A~2Dc%b=!7Hu#+TNpO$hs~`QL8Snl7pcVhia&xt-%d9=As34N1 zX5G62z-+TlQ)PS;y?#yk7Tg3tQ(Voo3~ z|EtRR7XB*7KfM6->VHW^|Cd+hKh^Se?DSV2v_iCi((!_}21!Dl+$;xBs3F8%t+h-4 zL0`ZGFb-~Qw4Td9zg&VOqIRX`F->sZ4**RSLFYdQeRK0zfQo#8^1%1{A)nH{eS85W zm~*1OJ19-aLI|p=s$#O%dG|kpbhH_2r#*^7Q(r137*Wr2N@$ z@Qadl)jjXF^`bp7pQd$<7IXQn0csJD1cth*C5UnJMf_|BBg!I`2_;|~_O@n$?^I6C z0eX%9`8Wii=wzuMeLzji5H2K+ieP+Ebcku_rR_~SjO??`X+gFaKm10;YovqRis>+9*mNW{s|3^22ML-k@-1S7W{nCCZpg7du-{bzZ#e)V&g(XnH zzN;h7L)leM7YHK-56qDD2qQB_@YF=r@}IO$PLCEi$bAWsxh4Eo#!S|nmhfJ{sLrm> zn_Tw?Z|UssnbT;or0KZsb)}=f{nPPb7!y%*4D4U1%E;^W#EJG5%uGuq{)5&i$}#7O zFb7*Yqv!inFA=ftN`vfu}IT>Wy8ek0=2e;Jbr<@a?9Z(OXS{d6ck4iShW;sZN`cCn3x#hetCKM@u{VZ zsPM_@D%D1d^JnRSvDjmS9Vspo5-A)uv;C#O8k2L${)>KS3=u50pkCn;A;?55uHx!_ zL3h5>vP*sze~D7h71{?-2fV=OU4^pNGy;Y}0ne_?3;Blw=+*t3I5HsMT&z*aqmRp3 z=9V8w1^5klJHzDc_E5y(jdX~JMj|=v)FP?xvF!RG;HhK z=YM`Q`X6x?{rf8Zry8~Xy`2B`<<#X?&s)ZtnVvdiBKgI+z9@thP6ON{*VLn>5=Vxx>$k9hbOoFVkhq(UF)z3IvFS&U^Dnj`CVp9W8B$FE`$@4DE6pCX>jz0_i7Yg+>1I|LRrX5>Ue2Ux`F>J5T??R?&T`>D~h zYwZQr6Oakf2skDa`QSNkC^%VM;JP)s)DD40^VvVQX;!GO+7M*BjEu&-0ln`-b7q~B{bymm;_>ZF&Qom~@n_shYba!zjx4?YUd-QNLmvD-P2e8+s=79(>FLwG1!}Vy>46BIu|{(_5Ic<6x_8`D0L3i0L`^dl zU>G6FY|fG$UxE~?;+hcA&)E6I<$}Lk+w-e+;L~e!dg1vffxZz*J_Quf2J+3`gIQTS z=v)0YD{!F@E0nrL64vPY&DTB$7yqdL<;2b1@0@QO>csVl!X(K#l9q8zo{tx zRc_!$4RtmZ=A_?}2b@_X3XZ=45mcCZ}b>$V`eZ0j@Oz@k7gF%~NIrqw)pJ;Ql__~-piIOQxhG0Jt0#gfG-*zl!1 z9`5m4h|rD}e`b;2%&4E+UIoD`8nt8Ra!?CtvU^ZVm71EVjP{XfZk>wfH8W8~=M(HU zNWWbd|G2qm+#J~Y2xeSTl4L<9R+Kap78|6*LxTz8{aD~bEB}5uWD9_)7(SUT1@bV{ z4Gj2B^wuU$fujm%*1_bjbVryj#os>e(Cj910I=BbcJH`!=@R_-0+~qQwg>T0Y!@PB z66ye(GK)-}!GD0yUM&0k!*D58{^rv5ST z1}jRoXt2>Y(MHtW1#awt8`=@cg0K0HbY9!Mre6J_fV6vg(qzp?*6!uWKU+QwQxx1G zk3!ioi!)?1;TF9;C3vZ`{+|qAmFJ}v5~nL)zkXO_x4q$~-K0z~i)d|dUmM7GMV@9d z{XclNt~dbR-ATCSHQ1!hDd3sb#(sP%Zb9+dVkN)!gDD}xokR|StJHTmAuc}N(1r^` z+RO_WC4xcm2`EdvuegcklP!nuM557(* znDzvfz}F<}pMD5q=X?_^qlw9`kX16rzdtw9jn(jlfN6=Bk5$F02BrA)+DM+wWjVgP zb*>#68!?B2k42~N|9v~;3AAF;dBW<+cf-=V)Gpu+qU>n1fCy$EfA{j1opr(ZicGkF zIrQRbHLuZi_PQn=LX&w4KmFq$^XASyBrEsA$hwKoX3AKp-?@FvTE=64)h?@TVP+O} z^Ek}t7*>+GK>GeBbTQq%g=HQZ***Beh{XX0b{C(EavbPWAQ)&nM zm3uPsZIa5HgcA9N{HexfT+mKkab`;+l58N8+~UzA0h}ZePQ0e&d zE6gHTj;|oU24$MSCtU_Pgjqy>jITeQf_v&`=5oj#$8Lk)cK2nq=?~zTGhW+|^wcTz z+myf48UVGwe*Bt0@^kB)=?TEO-13_L4i{?$Tz7vH=trH*!mEl?~ zD;Mr$@Gbgl~n`n{4MgkZZ{D z_5GvN4D#xnAg~{fT_%4@(D>y4<8swg+ga$KAioYM{ZTvagJaJ!4lbqOvU6;?{3$m8 zqSa`BAXC*yFE*S? zG&|XyMU-3*fT4YRMey40o4aN=kxs@+g=5!zpsk1k9<7jFIZ zu~F+wJTAmIR#SMy$8-PVvd{ZR_|0l883@9E`{7y&ylc!NGy$A{ts;28&6`?8cKmeY zlO@8}q20>kHYDE)$Ua6lXJk*0GYrewIQz0EUPgT;yyKgR-Yhf9a)f{Yw&HZeB^q1{ z{b%jD21Os-mCM2f?h`|uf#muvh=Ah?^4xuOuDnVVwdyeZgCvfhr;Zl}XF3e#$;s!e|m`(mT*gZ!``5WqxlCM8;2mLK;J+PGsfHI5p3&yJx(O{$i zbNdBx^8PTQx^VgV=qjTYkw7L$;(dJpa)i~hU1crHN?z>~4JVLwzm0~tM^{Is1V55P=z*HRpkh_=Qb3g{;>5<_?G?QcS&XR(o)XbmLn z$l^n33Xb6H(Izjar>iZO{h*!j4(FggLt--9xzO-s{#3J;A5J@K_wLgf&VW$;5^tFD2V8*7ERjC{{-7<7k4~bFC7=)_N}I;D)9N zzig9S*W!kYa|%P<9U?)7LJFwVHUW`Fn9EQe#qTCMFykBWY~9iSsqYSxJjc#+1l z%OZhHh*t&(K5IdLXWWa6V~zGT1~NfPpIp3v(LW0cpOKx(uDE3~!Y)`eb@n72pT^N6 z5GGk{Z)l)We~W2eRX^<8z@l~^itG}0xsZDR3`1KzZg8CU$m8q}xuI1U7O&I2n zWTb3hp@KaWr*mF)3vHR5)e_Eg9rlO*)`$W(sdu*ktmp*onnMTBoA1?J><0fnuf_-3 zYneruU9e4Z_C`F3&3~PZ!-Y6U<>U;axJ=Dv(pl30bI?6FY--M=PD?1ICjSwd{oUXe zrNT)T6*032)O3My7{wTp=GktKTgY(>P`l?Nb~+e@(BS|vvm{(;J926n482!STuaqj zl&g%E=<1sUK}wQB&>`66gf@%ie|xU9TLPKX1c|ylyCPE{ML| zo`_jmK%F2<|M+q=zP$eOj;~n*fIcc{Z#`_qMHD21j0PM2b+b)>?Wpeyz0j%_i-Q1z zi{1#nLm7=I@Qc7QCkq3}fk;&qIXO86bcUVZ^fA1MXGs4Ti=1$j)MH_%lC7jAWFoB+ z13=M+EbnBbN4D^QJu@_mT29D6@%N0*l0ojbz5nPFG6Bipv6O_$NsKo8%{cZV285C& zguER;MAN2W=tTcUA~gI>urdY0wPrzWkmv#-*Mfaxj`=eYw-Uc+j?yoNmb;96u5F(i{kWi1Yn7vb1tg^lotI@M=Mdp{MZk^b>Ny#5qtByT%9 zYHgReWs#Gk{v#Bf#KC_KbZYds4LXWmhcYVAnF=sxYNL4rC?Tsj!;8f83YORgRv#R^ z%4nbaZ}wctYcGu7%xuX++fN{KrK$T- z@t_y#Ts`rin~-u?6>m403BJit@B6|D1RN%ddiQj#o2%?m_5A$@+E`n&nK&p3Rlexu zWU*mT|9ePlgeJ>oA`O^CkH9yh#iriOzrQ{J>4sUf9fnPGc{)ei0TzX!T41t15n5;b z_oIPkH5Q7Z0kmfDs22);)5afj(R(UR_X>rr0v$NyulxJE=Q?+?&0qm#vB??MnB!6m zr))K6`k!BDKHnH-%MN#~-wZVRJ)8PH>rfY6_;X7y|Ga-=5tTgHTGS`bVw1JmqmguG z_{LRO+Q@3uNvHtZ<^^%V9*V+2aO^-V(lUkrhw4Q^>lgXAJW!QYBpGG(z;8lD2SGne zIIyJuzAZUQyGBuC7onj}d+3(n0n}_hqwz}c(as0;TUx>?6~O3c4&-C^N5vnvd`*8D zOhv>T*#ir-n*-V@y!q!Bq}uktVbTec9Kj-l*X0HR6>n$tX1v$!I2rFhy#VMEM^D^U zba_fqd)wduLZ=Ki@Su#q$JP!5hdNr(RWG zOqS|!oEEP$Hj?@|*-SL7nL#H!{P%|m%ut5@nu-jYT8bys5yfuvo9^(Vea=Mp-~kFB zq}Dc1wr>vO4$>4*n0 zix4s*G`2TxBNQJfXruo7qEXKUxq{hn&E4_nYoruM?O{&2^swP{X0eUI;nGBezm_h; zZ)RfxO%y=#oo3^t$pSAPK6W91aNjKYzsFu!xhb`XUHlx{QRG1VfhN2pnYoN zuc((^!dMLnuf21PU@w0-vhc?3qq+;+1oeQPS=c8W!RHfI|Cj>7`olS*{wEKn{Rbt$ z{$mQnwLi)9yMIi9*zl*kvH9O`@h_}}b#u+U^h{XoCFu-OyDBP~yXdp0_Am0_Mmu$s z*_o`P$FmbW$Wjh*dYb4LQ0_0BgVL`Qi2NpABOs)pCS0f1#iad$TIYnEuI#B%M9dOm zUEr`lqq!qXb>M0MlOUihymG`663_o3Dn3*UDWLideSUAD^5v0BW ztqv#Fl}^AupQZb&V8!M2iR5w~-#6;{`&FcPG5G8Js6C*kN*U@&0b_-##I8F7G2JIx?E5R@uowaVc~9}G(I-T-%0Qn8>M-*g~1Qz|Nh-1WtvuTv33KNZ9o zf0tdl7hCM0Sd)>LNfL1_VSa&W5cTOFM+!}MiD-v~eW#@mivgGXjV{XmABNpS|(VMaJe{UwM&vKdN{V*J{ z&dy{u!{!Kx9Qpu$44IGQVbuTW`7*Z>`XpkB&MRGNOxIqzs#z(kAG|v z2%bqbL=*p-C*83`7>s1jZW4+@duXxo{W*o_Oh82#h1}mob1sZR7=>#df+w5srk$D=x&#IRigV?dO(jz%H(_3P^(;Am73hDLk`qj3SV2#ujm zps6guqSv9hBs4_8C0;YBE_pP82D%MowRS}(uKBJ8;2;6j!G?WJ73zeY?$>lJI}_qV z{YQa3eb{Z#!5bXA6vAW_xE4bU{6@WZ7kdPAqs%1&md;9aL zm2)FY(fQr|-{jZY-v5@mx|b)X#Kjfe-gCh)FH1xESKqT;)qu3=Hluy+c|euYvhNnhKRK5d zuz8n{&lGU&@>vei+C;3Qq`V2z>h2Q{M>wXhFhN*6TQ~}M7io;97?h5XiJ1 zc|Nrd46X(HF6w%lB${}Kk$k!ug}LCs#5_j7wUQ)A`1!~3@t{il#uj)UOa$Y~3LU$f z;EGZPi9pX16j)m3e!*S>8iWfPpl@NoH2A0anVi^U*sj43fqCOkiltjsspO)lYJy=U&T-d zDK*(Z7}8ybIZM>{K+OWwKSy)I_VNX|RyR6+Z~uzYPfhQwgSLKf8H_!f*h@JF^Yq;2 zI^X~ags}U`OFwG#|JT&tx+pZ)a&vx3y$&`TfFPZ%9{qW&8XkVj^fDlP!a>m1Ck}GLO8v+?=so0lwhRJ<= zvOthYg{pAX!5ii21|p1Vd-9hc&^m!}+omSTkt=(Yt1pj`F%Om15Ogxn}eLHFZFEu6dh- z4Q~N*?*$oDb}vz2;!2yuYe(=fEWk98a_ZFb^ZbogV`&V-a|s$PyA$&B^Pli&=o2_Q z8O$tc0NG9zI0=r>;S+fa15ql|AUVvnND^Ip#H0dcopHg-H61NA$^qnmD*N!hH4Ph} zqcU>Z84CNET*ii~s_kgbT9KMk)8kb$6cHB}983q7{BvZFLZ4$h4K&DcgzxlA@pGdjd*DSNZIqjH zk<^C6>^AVk`@w~CAFg4jCK~3>SYPsr!3rp1#-qVtH|~QBiAa7lDuCK9ygQVd4Pr`P zP{k}7uev5aT*3Wkp^V_YyiA)~DX2N6xR4Mq!D1}Zere9AO|CNIL?*hqnM?a--`C4{ zs&WXD>C@dHS`gwp;X$CrKlleM5%6xnNYY&}>&PXvrfGAr4V};|&t##i=JErW$GA&Z5!b~GVIol%=!LdxscOa0|0vhY+ zM3yPc6)WW*12ro63>sB=T?r}nrlSFn796qp+Q9;slOFTt@wpBFlIM`G^~ap%+_Rv5 zV9YC_*#MajeQ`AC8Yy1{^;zzj(jR!lH_>|%-yoZG09}BYh~dy+xc9|F$MfH}KJ#m7 zoh5qHf90k;(rj7uD0%G5Fh{B6&w@Z;Vg+?JzKOun4YE^G{@+};gAkP7uJv3|e z1{34FOj02%(-afZ<_2)}`3@@7vk!C-Mv5(Fz&(fK?Ug!zSyYY+4#rrQenBYBL>qhk zOqE6pwb2kI~uk-GkWZ;erBXxp4-@$damU|zH8R3 z8F+g^Iz~H=bclrXOJlA_du8TJAM}Lc19vnFKa2WFVQ$>WnntH;2N-j~lbFP$n-jiW zo&$YZd^%al`Qu$Zn2J=YX0rw&1d(JE{764`nxHO1h(K1Ps~JNvkVuTiG+h!X2C>){ zcobx)`If&XDu<@;(Se}MjZs3%-~crT?2k;)Y=wl$z?Ydj0z}cb&#Lflox%d;J#oe0 zKv5Dj?!xMfPgmwDPU!{$)T$uF8edi&e+Pmep$pAxWkqPw!UO{E4QLd!C1Ho$Cb*Ir z{>)UtfiQVVYQm<0eqyW@QpVSzreToJSIua!PlKo;SG)ZW-j(!3S+Mf*9#NYMbB<_? zpVEDB+bbpr&gT-6Jz!*1rU8gR;Alk+74o)?atAbs){oPv$Z}|A(yZCAka8SVqQj69 zGfL)Zvq=-`1SUcMSgo?Hd_afk}k8xy-F+CJz`BJAVFHonuwWKF;7yKej@OoJWhD?dRa zxWriUa5j?sTskOT13Lq0Rs(&%G;V~5>_;T8Y}t-G3JrGg4)6q#Q)dw3HV>`owwD5C z`L9T1ncWxHb1ttEc>8Do(G1P^Qotl#>RUI*2oMF#p&ZCA_p#5oym*;b;Ub+gC~Blx z!8gSEF%gj@*3>5}jJDL~V<&<`XQh^0nnrGT&n$wYqo73i3_`2EiHmMN&CeNSoQi2B zF|QOFX)vlH9%89Y(CEDT$tTuKjR!okWkWB~dvb0}xoO*vBce%k%h6YjHF|Lp`MesJ z?ixu}@R)6A`)ptSwlDZ{KeZGx`ImQo61B>vVr61{aNmt4%86s;YU?`4>jrapLH)B& z&YPeABubpd5U_$LGILqz7{}otgCxy&rn%9wNS?g<#R$qc@1eGCzTv`h}rzy6rn*6Z-fosX8DAA zGdZhH6}fq@(cAlKVQt>1J4!1IkAz{_s@Zppdm6H%Y*FeR-&vc==JzhrtA5dUgQBJ} z<~kyiVskBQ1XKf zkkm#|Z&`q2F5c1%rOOc_EExH|BKQQZ&_IPTv<(DDPW4z)8#2Hbhi26oC7(rG@Ns}W znhbUmUfT}no2KoS$&AFqmD%p*gR@57y*d^&GJI4uAm!cu=WVA)^%GLdgZGNh+!^f~ zor-X7TJ{;^Wvbk2ldwcz+Gwuj*U}p-ajHod!MFlTHWOM|fS@lQ&CO-jwX>G(E?jiY zv{G2Rw?4yYqwsZzv{E_mZj)7qwi`JAkN~)F>u`6Km@jk_ef1s~Q5Ni#+Ko1i>zHZt z713biJrs*@|C)qYnp-LuN&^E%aPFph*jXLXe$hb#8v5TnmUV1( z^Cq)+8-23dny;e|wjJ}LS!}@o8lvJ! zm!y)4DsN_~8q4I>9_>nFeHulBBy!_po`lZIt0xm7po(Ck4h9yWj8w=*+%p=zG+QSQ znk4zpXwta;4(S%mB^~{F)adM>yUmck&212d6r~Urtp6!>Uz5BpWok2%`|S%-hRHtB+C}K|xz6NIRNfY%gPeS~+Rw*r%q?Sm&Pe4{@V z-8I`Z%6ZM&b(Txv?_9v5-x08uJW=F=CHpD64Q7k)@vOtR&KPSmW*Muoa=ivM%g|); z(v+~h7w{-H(^P0G72sErOe@8y5#%m`M>HutAJI=O_NJq{V6bHVn@y^+6M0FB#rFBmF z8R|J8Ll(SS!K6?@uA%%z$Q=7!tjMt=FGit(4>%3-Ce~w&PW1Jq-X$H|UFZPa~@xneAu{Q|JFh1?*d33_)Oio_6plfhq75&vDI zwT?Q#HE^y?4KqezxGFymvlI=}tsT0q%$r)lB|O+tP_7cxXYsC6*l?+ zRwW+wsnLuaglfy*{<#3Z#bN&lGMO++hdKmK{kV3ZCiCBaSxWLE4T0`}@4p12xw0%o z=k5&Rg8FenH$+EYF29UA=>UK*I6;l#sXYpD-G%P8PdzwDuO4^xZ4QS|$7lEebqNM4 zlgEpSIFKzfS#Hu_$4u}<+8#Fi$L{rVuG-qy&e$4T9Np}-$98Y9 zqt)k!uBU3BU$8bSanp}mY*%G%yCGaQCOsLp=2+R}^;aKU*9BDHbW-K>$F|9Zx7THr zX9qzqPtV9OM?>$-=JRuwmB#BPID&$5c5&gMvKSlbHKP78Y-1tcxYy zNg+|y5=<&~!ld!^%*=M22|rHA1K$;soihhAlD8taZri@y5}j&R7@K_mZBfLZVDKp!iTF@t@?hTRrALd{uhGQuvOd3 zBUJXYk9Mq@<9AN> zM?hnWc}if0kP*GEs|n!&*5Zwx_+174T>2E2=G5nZw#3m(?Ooj5VlfE=#sL<)qdXEH zN8`WTT(8&T0itT{D0!`{tO(K*WdTy~D-jWqG0+hA_}gnd7>o}$a4$O>i zw+51X1(ypy@d$qB00L7}Q`6AMhy|#O)7@o}YLVc7VzuLlR#5O7v5ipnV1t8gN1CMS zKi_XA0+E*f`MAk4agf3IhJCiHJPS?UIXVpDa3 zeh22cAS?@~S{ym@Wl~a7w%4frMvc&GFgVBq$q|B-zd8%cbZ-w9+kZ{6XpeS%K!Q1Re5k@ zh>e}pw0CoJGcE89mp031vS_Oyr7+d{euL4{3f2BKH1D-xO*=8V5#ez@KVQYQ=P!6l12H-+7ytkO literal 16175 zcmeHu2~-nj+i27(ZWyU;si34vEnlTVl`6`TST|}_(5giR5?4?_qD9C6Nk%KnSE+#d zRYZiiw4z3cl}*f$Sg0&TKtcJ{$j&d1j1B+K$vo1`V_n-u-P`4K$yhZzIE3|{9gjWg+TBp zxVR8p@OKiyKhMd@39m9%7$1KZmn0W|e;0rJo#&F2+d@LNh2U@Q zwuG|1d-wW}jOFh_z`ObT6O8?gPZIv)pM<~j{F6lfd3eF_&c?3BCm|#OFWiuXvJfs_ z+7j?_L-2xECnRvW3EZ*-ys`{0Sdt4N315&uA;}+q8^<+{nv|E6hZj+j2rn2nWE(dj z#MqBp#w}~(wqbexggk$jyd=E%=i#l5k0?)s7c39M$`ZDfC4`i5@xm=D!{=!$Ys1Pg zOqApzO7a)w;U!6gcQE$A9}NEyVOSeCq>Y=lB;Nr~<9BLu<+gv}d1-vwOj z7wgaZGQuQsi_7QdJ`I7U^|{lrBhE&v4sdH_D5S1=6sJ%esad*FB*Bj4PSDmFb=Isbyu`ii@~1O z#+nL8DC&^@Cv{rmJQ@GKNxg=)l9}xA{-`+;v$bD&&xJ3MDyP+h2Mpb?u6fgG z*)tn3O8wKul)4u_Z=v0xlsew#qkkAn8Q$;t=ae`*h9XYQ{jF z3xo{M7g)}_yCs&fItHEG(p)TJC^h!+pH+d{G>5{xfsm<3FJu|c&y0W0?goX-ev(A> zYtIdt99^qQa+puuH*wrmHTNc2fHwvAy z3*X(J{WW8iOZ`T=?IGp&EKW2!`O2?rAV!tom{eTNk_K=c4S7)sjk$L9)uF4?kGqkA z)!P-1W;LEgJ`sox(e?>v98fzvcGwet!rYwp8KRK{>$vY>Ov56i+}hcaGITdT4&3c6 zyF?Q2ar?>u-?2Qpx6X{?1HHv*mOp#c?u;oy#a<6}9hrGH^dk|i{~A`h$P7G+@)GZq zq`mSy_@Z06W^l{H!3XLVKC|Uq`V+G!X>@@9jvsohNtSv!ec!aUZ_zAHd|g0*J(QXl z$VUq1w`6)0+Y_mKu-BES4G$DpIBuuaMRAegwgNMm-dWP9={V0hByTxx$bXG@230i~ z)kRT}!?)tl5dFnIx8TM9dVF>dr;Nf`Z9vrTZ*=On` zLB&~g8^hoU!&C9Qa4k^+%&RO6mAd0-()!@3w_m6E?6o#BdSJtRIFrKQO@8PK1&|7k z$1o(Mo7K$04&lkNrtVNtN#R}1(_QFBu=?gozC6UeGI01x2m`2Q*LEkmkp#;EY!0Ne4%XL*BmvE{a&}i6 zdObNaf4#V^P>AjRbVJ~(uSY@?a45D z;GReX9OAIk6Sqve%BbaV52)SZJRrwjU@MRz3_T9mH5_v-EnGVfF35}YDbBL2vx8iT zuB;2dV#lN>?=r(5Rx)^D^dUcYv%u2q{8dX;U1&PikduwIEcdb#0q36J*dW8|B|)Y0 zRyn9TS{rl0X8Gl58K`V^Umv(}y5sXlT#N}7+xe{?l+Vnd|M1AkdLCR(S}T_hx)!cA8Dpz|i1R2MR~o zVy}o{7&a;A3$UXC?6|qBFcVYry;caa22ui|-3O-b6V%_++Uf_6+XaAzc4shTF#9ft zt)Ii)d0*-dHZTpamnN=*!;B&NO=)5*krpm?B{!;@V4;WVb!{~$eV=4d889eV9K7Fz9;-P&N z#^Ez|x@Q@nxkMZ&RVacC7qWmsHDi_kaMyB4nd=_Zbl!*Soqk`~nLvym45v&Mec zQj=9~9Kp@eExz|V0*|);9Ti`ZK6BltSHy@`$^bUFc?--%-ab0(<6DElT@ z$M4+#FWIB_d@8}r$obP-<^Qbz0#^t|1;OuH!<|1Qz~T}4fBg^X5kI0N{K@Hm*CZx7 z{w{?lfHR_RPGHU-de8)Q7)^%>*71je{Wth34;6sDjg0AI6!dFFTC~=-NMx#NXP~hA zV#po0@a7DJZL|4l`iC#8k;oeHsacBzDo>iUiqMJ+NRW-Cj$b>m6Cj;f_9h2rEHNQmjU zOCdW)##6s`WpQVzAoG{*cbC3v+V@@fNDzg|R{VMmm1UUFX!3M~@9gjRHtb=#kR4J# z!Ck2*_7N2>;TwHNqrB(*OhgALHYiWUW<7G_9x(c&MmZI}yeR!VAi%N8OW@=~mMww! zxQlAk4N&(8twK(TSvNs5QLd#&c)l)3<6ye}70{rWL|@v7bve875 zEs15!%<#Um>wn8ZGCg9Hh2R=w*;u$mZUFF|hue_Wo?LFS-m#i+uZ7fIUFj7?Nw3Ue zNl9!?jZph59^_z`cx3sRs?#2MJlIpKI|NW3`-#&&Wwv*vkJx_&}^j6mRb~Lx>$S&S82oB z0TBho88HjS9BT|iTQl%B96uIY;FQWNP@WOQB#-@a2Th0EG{w;~=tEmhr;P$)Qw*c` zQrldZWr7?*hXD^Pb6&KwX2}!JQ6utJ5r6o*=QEHoC$|(DbWK*x#yw@`_x`YxaBP0M z9(J4uFE<@=7vLq(+GRK9da>B;v)t19)|u{ZDI_9MelgqLcf*9h#tK24em49aId*;y z5=ALklj_GWr|Gjfrj%%oEHpf<*kQW7!vJ4KR@|+|%IzF>Qr|=cN#XJ+ouxk626D63 zM_7+LKAioyr@IRbM{~C@L*D4U3&^HW8Z~gTNCC^5Lb78$H;DZ4#_CPsJ0rs|E5`F+hnXa-7h50Wqagmtj%-setT(_zQfg7ML0JGg?)j>Wn$a(LZ4^1IBpK2pd7<*u@6d1>p<@8qZ`Zt(Kn~y#j*1iZ{rBELp2M3&o zJJ@to`PYZQP$Mhh;m<^%LU^ zS^>qGjye-5uMpdn?~O_Jx|ltNkyHJt#bV?VdHw8{|pk>ak`CkCwhf_T}R zpwMyX_bH2zuK$&s#Y1-?4)?STmxS+&-w+pFsONalkM-k+6`<+5jtq(|G=^LoP_jyt z-#JL6U8(f1FD0kme^Uow_>0ZF7F1v6%t=G3reYVT5-LZ8@7V^_{xW!rv%O>d>CZw= zV(R&_VNy;>XJCVe31Z{IL9GV7)$Q4r`frWpT!x$^eMae`u?t;Nq?HZOea0>S4(cFz zbs|6=Q4UzYAwF6g=ZS0PKW-VgA6qyUYmjA><&E)!myl;F=GbW-p$bkSL!H0a*wvN7 z6L5%ByPtk~{@|6bx0N@YMsJo*{5=7O2^^Tff&UIT@Wd7(3PC)nlFOS1Kf_Z=XV={O zc(3(pW1KL-g=1JL-S+(ftEZo4UcH!mW<5YBxob7Lv)+m6tT?pePpkr?0+$d-u+V;TOX3n>>rRO5L^5D{9uuf5h$WkyP351AuBW=Mzny z(dGO`4r~WX#d9Yw{+Sc2z^{A=dv$K`zx%sG-a{9HHvgGe8--_@Mh-RFt9;E(D?3TP zg>?O94<#&Q*-{QWa{~4*be=ir=_XzMV zOV|3saZ>m}qmYb4t>l5P{$V`lQ`31CyiK^Q{J*RIPhI0eRvQhYp-ylZQ-IZc4;tlJ z{D)nJ7oo1~GYs81jN{|+?8FID;ZQe?gPj{3yr!)jGGw+>D1$%L=d=3?I1lC&xmKZ% z9|HZVIvCju@TwY67zsZZyPkgvs28TxN&Qrcj>ILD(J18%FPxzB_#$(n{vYx1q#GDkFRK4#f{<)4&sx=PPoL&n@ z=~JdQ>?d$#0XVc(d7_&kM#N9K6(ALA(Fwq3wT0=dp_hjIDXkY}3?wE3&>md~d+5$N zf;6!5)O)jG&p0`w+s@txJW(1uLHJ%4c)$~Gh!%@01 zKpS>M-d@Y~#yDMx)Fyym)G{0_{{cGxqISSrE9Lg0bDG z=7sf6u}s-C9Vy9HR~Q^9N$yf{)BaNW?P1a?PwZdofaMe{p+Q!8UK~`0hBU-sgQ*pz ze&T#stY+@q8fLISma*bXS`%}jkOguIBO*fD=1g`W+IAd0icrw)%S4g?(5)h z6`GaZY~__8y`9uB95CEBAKua0-Ir};BexegLHb>t#e7v&1*zQsQYINVqPBmeEA|l9 z7N8E!tF@`$Ij?-ZxYzqk9Hyx>Z2q!>;u=qYENShYqvono|=nOrb zntcsn-C_xVv<)0v%rVdc@atBGjL5nQkEsnzLJux1idWZV%=Odn=u%2nyJ()}e=THa z6BRV}2b!L3N%;}jRjx$em{MK)V9=Wx0Oa3p!qK2JQ1yA4-$n?l+_*uQr+Fkn7^LYZ zL2dWp5sLdPT4_%K0z839A8e9#g->@ptuS7Meb7SsWg9RlrvuL-%9U)!FLUN}+)CEi z-G8<5KX*HJQ7fQOiC#EPdA(S4CTPTTlAqh@olS$_?ni!CQb;eM#K4i1lZ+Hd=(W_ zGYuk3|7=T!-buRAe!AmJpQ*#eo!$qG(Zru==~zHfy#HM4kNGou(i(G-Wx^J*wb;j| z%7C|a`n@%veiRN<>R2s0$~@|4XVJxlN49%WZ5}s)@;??|)3HBR z$**UTV`b+bF|($9h=W67?fjm!|LS91@L{jQTtjTXai*z%?*{Up*Mt7*n)08G^k6}< X&*X=-sh5ox-#2gc+>o_y|F{1E_Zgj! diff --git a/docs/articles/06-vs-dask-juliadb_files/header-attrs-2.8/header-attrs.js b/docs/articles/06-vs-dask-juliadb_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/06-vs-dask-juliadb_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/07-glm.html b/docs/articles/07-glm.html index 75ef5ee3..d208273d 100644 --- a/docs/articles/07-glm.html +++ b/docs/articles/07-glm.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
                  @@ -39,7 +39,7 @@
                • @@ -95,7 +98,7 @@
                  • - +
                    • @@ -109,7 +112,7 @@ -
                    +

                    Generalized Linear Models (GLM) including logistic regression with disk.frame

                    @@ -122,15 +125,16 @@

                    Generalized Linear Models (GLM) including logistic regression -
                    suppressPackageStartupMessages(library(disk.frame))
+
                    +suppressPackageStartupMessages(library(disk.frame))
 
-if(interactive()) {
-  setup_disk.frame()
-} else {
+if(interactive()) {
+  setup_disk.frame() 
+} else {
   # only use 1 work to pass CRAN check
-  setup_disk.frame(1)
-}
-#> The number of workers available for disk.frame is 1
                    
+  setup_disk.frame(1)
+}
+#> The number of workers available for disk.frame is 1

                    GLMs

                    @@ -143,9 +147,11 @@

                    Introduction

                    One can fit a GLM using the glm function. For example,

                    -
                    m = glm(dist ~ speed, data = cars)
                    +
                    +m = glm(dist ~ speed, data = cars)

                    would fit a linear model on the data cars with dist as the target and speed as the explanatory variable. You can inspect the results of the model fit using

                    -
                    summary(m)
+
                    +summary(m)
 #> 
 #> Call:
 #> glm(formula = dist ~ speed, data = cars)
@@ -167,21 +173,23 @@ 
                    
 #> Residual deviance: 11354  on 48  degrees of freedom
 #> AIC: 419.16
 #> 
-#> Number of Fisher Scoring iterations: 2
                    
-
                    or if you have {broom} installed
                    
-
                    broom::tidy(m)
+#> Number of Fisher Scoring iterations: 2
                    
+
                    or if you have broom installed
                    
+
                    +broom::tidy(m)
 #> # A tibble: 2 x 5
 #>   term        estimate std.error statistic  p.value
 #>   <chr>          <dbl>     <dbl>     <dbl>    <dbl>
 #> 1 (Intercept)   -17.6      6.76      -2.60 1.23e- 2
-#> 2 speed           3.93     0.416      9.46 1.49e-12
                    
+#> 2 speed           3.93     0.416      9.46 1.49e-12

                    With {disk.frame}, you can run GLM dfglm function, where the df stands for disk.frame of course!

                    -
                    cars.df = as.disk.frame(cars)
+
                    +cars.df = as.disk.frame(cars)
 
-m = dfglm(dist ~ speed, cars.df)
+m = dfglm(dist ~ speed, cars.df)
 #> Loading required namespace: biglm
 
-summary(m)
+summary(m)
 #> Large data regression model: biglm::bigglm(formula, data = streaming_fn, ...)
 #> Sample size =  50 
 #>                 Coef     (95%     CI)     SE      p
@@ -189,34 +197,35 @@ 
                    
 #> speed         3.9324   3.1014  4.7634 0.4155 0.0000
 
 
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
 
-if((majorv == 3) & (minorv >= 6)) {
-  broom::tidy(m)
-} else {
+if((majorv == 3) & (minorv >= 6)) {
+  broom::tidy(m)
+} else {
   # broom doesn't work in version < R3.6 because biglm does not work
-}
-#> NULL
                    
+}
+#> NULL

                    The syntax didn’t change at all! You are able to enjoy the benefits of disk.frame when dealing with larger-than-RAM data.

                    Logistic regression

                    Logistic regression is one of the most commonly deployed machine learning (ML) models. It is often used to build binary classification models

                    -
                    iris.df = as.disk.frame(iris)
+
                    +iris.df = as.disk.frame(iris)
 
 # fit a logistic regression model to predict Speciess == "setosa" using all variables
-all_terms_except_species = setdiff(names(iris.df), "Species")
-formula_rhs = paste0(all_terms_except_species, collapse = "+")
+all_terms_except_species = setdiff(names(iris.df), "Species")
+formula_rhs = paste0(all_terms_except_species, collapse = "+")
 
-formula = as.formula(paste("Species == 'versicolor' ~ ", formula_rhs))
+formula = as.formula(paste("Species == 'versicolor' ~ ", formula_rhs))
 
-iris_model = dfglm(formula , data = iris.df, family=binomial())
+iris_model = dfglm(formula , data = iris.df, family=binomial())
 
 # iris_model = dfglm(Species == "setosa" ~ , data = iris.df, family=binomial())
 
-summary(iris_model)
+summary(iris_model)
 #> Large data regression model: biglm::bigglm(formula, data = streaming_fn, ...)
 #> Sample size =  150 
 #>                 Coef    (95%     CI)     SE      p
@@ -226,21 +235,21 @@ 
                    
 #> Petal.Length  1.3136 -0.0539  2.6812 0.6838 0.0547
 #> Petal.Width  -2.7783 -5.1246 -0.4321 1.1731 0.0179
 
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
 
-if((majorv == 3) & (minorv >= 6)) {
-  broom::tidy(iris_model)
-} else {
+if((majorv == 3) & (minorv >= 6)) {
+  broom::tidy(iris_model)
+} else {
   # broom doesn't work in version < R3.6 because biglm does not work
-}
-#> NULL
                    
+}
+#> NULL

                    The arguments to the dfglm function are the same as the arguments to biglm::bigglm which are based on the glm function. Please check their documentations for other argument options.

                    Notes

                    -

                    {disk.frame} uses {biglm} and {speedglm} as the backend for GLMs. Unfortunately, neither package is managed on open-source platforms, so it’s more difficult to contribute to them by making bug fixes and submitting bug reports. So bugs are likely to persists. There is an active effort on disk.frame to look for alternatives. Example of avenues to explore include tighter integration with {keras}, h2o, or Julia’s OnlineStats.jl for model fit purposes.

                    +

                    disk.frame uses {biglm} and {speedglm} as the backend for GLMs. Unfortunately, neither package is managed on open-source platforms, so it’s more difficult to contribute to them by making bug fixes and submitting bug reports. So bugs are likely to persists. There is an active effort on disk.frame to look for alternatives. Example of avenues to explore include tighter integration with {keras}, h2o, or Julia’s OnlineStats.jl for model fit purposes.

                    Another package for larger-than-RAM glm fitting, {bigFastlm}, has been taken off CRAN, it is managed on Github.

                    Currently, parallel processing of GLM fit are not possible with {disk.frame}.

                    @@ -262,7 +271,7 @@

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/articles/07-glm_files/header-attrs-2.8/header-attrs.js b/docs/articles/07-glm_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/07-glm_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/08-more-epic.html b/docs/articles/08-more-epic.html index d8200058..7e9e386d 100644 --- a/docs/articles/08-more-epic.html +++ b/docs/articles/08-more-epic.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
                    @@ -39,7 +39,7 @@
                  • @@ -95,7 +98,7 @@
                    • - +
                      • @@ -109,7 +112,7 @@ -
                      +

                      {disk.frame} can be more ‘epic’

                      @@ -132,52 +135,57 @@

                      How to make {disk.frame} more ‘epic’?

                      If you follow this post by Bruno, you will end up with a 30GB CSV file. But this 30GB file was made by row-appending 303 files (size ranging from 70MB to 143MB) together. I will you a few ways to handle such data. As usual, let’s start by setting up {disk.frame}

                      -
                      library(disk.frame)
+
                      +library(disk.frame)
 
 # this willl set disk.frame with multiple workers
-setup_disk.frame()
+setup_disk.frame()
 
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
                      
+options(future.globals.maxSize = Inf)

                      Load the individuals files

                      The best way to use {disk.frame} is to load the individual files, as csv_to_disk.frame allows a vector of files to be used as input:

                      -
                      path_to_files  = "c:/data/AirOnTimeCSV"
-list.files(path_to_files, patter="air", full.names = TRUE) %>% head
+
                      +path_to_files  = "c:/data/AirOnTimeCSV"
+list.files(path_to_files, patter="air", full.names = TRUE) %>% head
 
 #> [1] "c:/data/AirOnTimeCSV/airOT198710.csv" "c:/data/AirOnTimeCSV/airOT198711.csv"
 #> [3] "c:/data/AirOnTimeCSV/airOT198712.csv" "c:/data/AirOnTimeCSV/airOT198801.csv"
-#> [5] "c:/data/AirOnTimeCSV/airOT198802.csv" "c:/data/AirOnTimeCSV/airOT198803.csv"
                      
+#> [5] "c:/data/AirOnTimeCSV/airOT198802.csv" "c:/data/AirOnTimeCSV/airOT198803.csv"

                      However, there is always a danger with importing from CSVs, especially when importing from multiple CSVs. The chief danger is this: CSV do not encode the type of the columns. Hence the column types need to be inferred. However, when inferring the column type, we only use the first few thousand rows. This is fine most of the time. However, there are times when the column type inferred using the first few thousand rows isn’t compatible with data that comes after. That’s why you will see an error

                      -
                      Error: Column `WHEELS_OFF` can't be converted from integer to character
                      +
                      Error: Column `WHEELS_OFF` can't be converted from integer to character

                      when running the below code

                      -
                      list_files = list.files(path_to_files, patter="air", full.names = TRUE)
-system.time(flights.df <- csv_to_disk.frame(list_files, co))
                      +
                      +list_files = list.files(path_to_files, patter="air", full.names = TRUE)
+system.time(flights.df <- csv_to_disk.frame(list_files, co))

                      I need to come up with a framework to help with these types of task, but for now, you should use the colClasses argument whenever possible to set the column types. For example,

                      -
                      list_files = list.files(path_to_files, patter="air", full.names = TRUE)
+
                      +list_files = list.files(path_to_files, patter="air", full.names = TRUE)
 
-system.time(a <- csv_to_disk.frame(
-   list_files,
-   colClasses = list(character = c("WHEELS_OFF", "WHEELS_ON"))
-))
                      
+system.time(a <- csv_to_disk.frame(
+   list_files,
+   colClasses = list(character = c("WHEELS_OFF", "WHEELS_ON"))
+))

                      which yielded

                      -
                      #> csv_to_disk.frame: you are trying to read multiple files.
-#> Please use colClasses to set column types to minimize the chance of a failed read
-#> -- Converting CSVs to disk.frame --
-#> Converting 303 CSVs to 36 disk.frame each consisting of 36 chunks (Stage 1 of 2):
-#> 
-#> Stage 1 or 2 took: 00:01:47 elapsed (0.170s cpu)
-#> 
-#> Row-binding the 36 disk.frames together to form one large disk.frame (Stage 2 of 2):
-#> Creating the disk.frame at C:\Users\RTX2080\AppData\Local\Temp\Rtmp6VQrsK\file1a1c77035a94.df
-#> Appending disk.frames:
-#> 
-#> Stage 2 of 2 took: 00:01:01 elapsed (0.340s cpu)
-#>  -----------------------------------------------------
-#> Stage 1 & 2 in total took: 00:02:48 elapsed (0.510s cpu)
-#>    user  system elapsed 
-#>    0.51    0.32  168.72
                      +
                      +#> csv_to_disk.frame: you are trying to read multiple files.
+#> Please use colClasses to set column types to minimize the chance of a failed read
+#> -- Converting CSVs to disk.frame --
+#> Converting 303 CSVs to 36 disk.frame each consisting of 36 chunks (Stage 1 of 2):
+#> 
+#> Stage 1 or 2 took: 00:01:47 elapsed (0.170s cpu)
+#> 
+#> Row-binding the 36 disk.frames together to form one large disk.frame (Stage 2 of 2):
+#> Creating the disk.frame at C:\Users\RTX2080\AppData\Local\Temp\Rtmp6VQrsK\file1a1c77035a94.df
+#> Appending disk.frames:
+#> 
+#> Stage 2 of 2 took: 00:01:01 elapsed (0.340s cpu)
+#>  -----------------------------------------------------
+#> Stage 1 & 2 in total took: 00:02:48 elapsed (0.510s cpu)
+#>    user  system elapsed 
+#>    0.51    0.32  168.72

                      As you can see the the 303 files are converted to a disk.frame in about ~3mins, this much faster than Spark.

                      @@ -187,67 +195,72 @@

                      Starting from {disk.frame} v0.1.1, we have something to help with this. When you call csv_to_disk.frame with the in_chunk_size argument, then the big file will be split into smaller files using {bigreadr} before reading. This happen automatically.

                      Splitting up a large file might take a few minutes, but once the file is split, we can use multiple cores to process the split-up files in parallel! The performance gain there might offset the overhead of splitting a file.

                      As mentioned, the key is setting argument in_chunk_size to a value, which will activate the file splitting

                      -
                      path_to_data = "c:/data/"
+
                      +path_to_data = "c:/data/"
 
 # read 10 millions at once
-in_chunk_size = 1e7
+in_chunk_size = 1e7
 
-system.time(csv_to_disk.frame(
-   file.path(path_to_data, "combined.csv"),
-   in_chunk_size = in_chunk_size,
-   colClasses = list(character = c("WHEELS_OFF","WHEELS_ON"))
-))
                      
+system.time(csv_to_disk.frame(
+   file.path(path_to_data, "combined.csv"), 
+   in_chunk_size = in_chunk_size,
+   colClasses = list(character = c("WHEELS_OFF","WHEELS_ON"))
+))

                      which yielded

                      -
                      #>  -----------------------------------------------------
-#> Stage 1 of 2: splitting the file c:/data//combined.csv into smallers files:
-#> Destination: C:\Users\RTX2080\AppData\Local\Temp\Rtmp6VQrsK\file1a1c22af50f3
-#>  -----------------------------------------------------
-#> Stage 1 of 2 took: 00:02:06 elapsed (00:01:15 cpu)
-#>  -----------------------------------------------------
-#> Stage 2 of 2: Converting the smaller files into disk.frame
-#>  -----------------------------------------------------
-#> csv_to_disk.frame: you are trying to read multiple files.
-#> Please use colClasses to set column types to minimize the chance of a failed read
-#> -- Converting CSVs to disk.frame --
-#> Converting 15 CSVs to 36 disk.frame each consisting of 36 chunks (Stage 1 of 2):
-#> 
-#> Stage 1 or 2 took: 54.0s elapsed (0.120s cpu)
-#> 
-#> Row-binding the 36 disk.frames together to form one large disk.frame (Stage 2 of 2):
-#> Creating the disk.frame at C:\Users\RTX2080\AppData\Local\Temp\Rtmp6VQrsK\file1a1c680f2305.df
-#> Appending disk.frames:
-#> 
-#> Stage 2 of 2 took: 36.0s elapsed (0.200s cpu)
-#>  -----------------------------------------------------
-#> Stage 1 & 2 in total took: 00:01:29 elapsed (0.320s cpu)
-#> Stage 2 of 2 took: 00:01:29 elapsed (0.320s cpu)
-#>  -----------------------------------------------------
-#> Stage 2 & 2 took: 00:03:36 elapsed (00:01:15 cpu)
-#>  -----------------------------------------------------
-#>    user  system elapsed 
-#>   75.79   47.89  216.42
                      +
                      +#>  -----------------------------------------------------
+#> Stage 1 of 2: splitting the file c:/data//combined.csv into smallers files:
+#> Destination: C:\Users\RTX2080\AppData\Local\Temp\Rtmp6VQrsK\file1a1c22af50f3
+#>  -----------------------------------------------------
+#> Stage 1 of 2 took: 00:02:06 elapsed (00:01:15 cpu)
+#>  -----------------------------------------------------
+#> Stage 2 of 2: Converting the smaller files into disk.frame
+#>  -----------------------------------------------------
+#> csv_to_disk.frame: you are trying to read multiple files.
+#> Please use colClasses to set column types to minimize the chance of a failed read
+#> -- Converting CSVs to disk.frame --
+#> Converting 15 CSVs to 36 disk.frame each consisting of 36 chunks (Stage 1 of 2):
+#> 
+#> Stage 1 or 2 took: 54.0s elapsed (0.120s cpu)
+#> 
+#> Row-binding the 36 disk.frames together to form one large disk.frame (Stage 2 of 2):
+#> Creating the disk.frame at C:\Users\RTX2080\AppData\Local\Temp\Rtmp6VQrsK\file1a1c680f2305.df
+#> Appending disk.frames:
+#> 
+#> Stage 2 of 2 took: 36.0s elapsed (0.200s cpu)
+#>  -----------------------------------------------------
+#> Stage 1 & 2 in total took: 00:01:29 elapsed (0.320s cpu)
+#> Stage 2 of 2 took: 00:01:29 elapsed (0.320s cpu)
+#>  -----------------------------------------------------
+#> Stage 2 & 2 took: 00:03:36 elapsed (00:01:15 cpu)
+#>  -----------------------------------------------------
+#>    user  system elapsed 
+#>   75.79   47.89  216.42

                      As you can see, this takes longer than loading the individual files approach. I am working on a guess the in_chunk_size feature so that in_chunk_size="guess" will work in the future, so the user doesn’t have to input a manual in_chunk_size.

                      Load one large-file (no splitting)

                      Split up the file will effectively double the amount of disk space needed. We have the original file and we need additional space equal to the size of the original file to store the split files. This is not always feasible and that’s why you can specify the LaF backend which was shown in Bruno’s post but repeated here for completeness. The problem with a LaF backend and reading one big file is that it can NOT be parallelized and hence it is much slower than the other approaches.

                      -
                      path_to_data = "c:/data/"
+
                      +path_to_data = "c:/data/"
 
 # read 10 millions at once
-in_chunk_size = 1e7
+in_chunk_size = 1e7
 
-system.time(flights.df <- csv_to_disk.frame(
-  paste0(path_to_data, "combined.csv"),
-  in_chunk_size = in_chunk_size,
-  backend = "LaF"))
                      
+system.time(flights.df <- csv_to_disk.frame(
+  paste0(path_to_data, "combined.csv"),
+  in_chunk_size = in_chunk_size,
+  backend = "LaF"))

                      there were multiple warnings

                      -
                      #> Loading required namespace: LaF
-#> Warning in FUN(X[[i]], ...): Unsupported type 'logical'; using default type
-#> 'string'
                      +
                      +#> Loading required namespace: LaF
+#> Warning in FUN(X[[i]], ...): Unsupported type 'logical'; using default type
+#> 'string'

                      and the timings were

                      -
                      #>    user  system elapsed 
-#>  591.32   26.94  599.08
                      +
                      +#>    user  system elapsed 
+#>  591.32   26.94  599.08

                      As can be seen LaF works and can detect the column types automatically. However it is slower than the default backend which is data.table.

                      @@ -261,22 +274,24 @@

                      Can {disk.frame} be even more “epic”?

                      Well yes! We can actually speed up the group-by operation that Bruno did by using srckeep. The use of srckeep can’t be emphasized enough! It works by reading from disk only the columns needed for the analysis, and hence disk IO time is (drastically) reduced!

                      -
                      tic = Sys.time()
+
                      +tic = Sys.time()
 
 # doing group-by in two-stages which is annoying; I am working on something better
-mean_dep_delay <- flights.df %>%
-  srckeep(c("YEAR", "MONTH", "DAY_OF_MONTH", "DEP_DELAY")) %>%
-  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
-  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE))
-(toc = Sys.time() - tic)
-#> Time difference of 2.800005 secs
                      
+mean_dep_delay <- flights.df %>%
+  srckeep(c("YEAR", "MONTH", "DAY_OF_MONTH", "DEP_DELAY")) %>%
+  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
+  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE)) 
+(toc = Sys.time() - tic)
+#> Time difference of 2.800005 secs

                      Compare the above the to timing without srckeep

                      -
                      tic = Sys.time()
-mean_dep_delay <- flights.df %>%
-  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
-  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE))
-(toc = Sys.time() - tic)
-#> Time difference of 15.62312 secs
                      +
                      +tic = Sys.time()
+mean_dep_delay <- flights.df %>%
+  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
+  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE)) 
+(toc = Sys.time() - tic)
+#> Time difference of 15.62312 secs

                      On my computer, with srckeep the timing is only ~3 seconds and without it is 16 seconds!

                      @@ -307,7 +322,7 @@

                      -

                      Site built with pkgdown 1.5.1.

                      +

                      Site built with pkgdown 1.6.1.

                      diff --git a/docs/articles/08-more-epic_files/figure-html/unnamed-chunk-2-1.png b/docs/articles/08-more-epic_files/figure-html/unnamed-chunk-2-1.png index 9096f2a83d4220b84823eef765db722b91527679..2a7fd4ee27a5f44969fda398c7568358f547a0de 100644 GIT binary patch literal 38731 zcmeFa2UL`2yEQzPXzU@zN{cvvf&-2{ny7smAzHV=vR5`w3 z?as9n3S|TBXN6M~%2!Sl%8IjJufpHxFRwPhw^iqqjw|3hzGjYOdE(oeOFy5nqfj`n zlmD0bb&r}*DBn|P3O{N%g$}eh-`i3fCf_dHa!`7u0^4^#t`SiC@zS3^dhGF4(feWN z=F2q-M>KA&wAWiD_Pxyw!JWJp^d5D5t9SUfPGbd|qZ&UP5IEMnWsAh5W=P~i#?cm#epC505wZ#<((SGI}-IVuu!+W9@cQY4nw>`wI5gZri~6>B0Gx*{mOS@V0+N zet5TMnK1c&_}Ch!W#k7D#aHXd_dnE@lc%G2-THQ9Ir+i(uGcp5{rkuNiH~RHX;eTN%Mi^EHa!%sDvdfo^5t4Qr#n9p5WoMX0xN#tDU&U(1oqPTJno<9Vd8*XqPv4Yy8bxR=W^?K38^MOfz=4!89VFs!eOZ&P(rM&!xH1>WIaIizCpRweuljUHkV{N}A8K=}%>)^gE;mixZ zjWQ#z?ulFKlru_#gp+D3N}oQE`!3x^tJTqj3~!)Y5sCI{M?1d@Yi4QcIa< z=J#2u^j_I!=yt$Ysz3U4vN5fkVV$?&ki9V3WHXpjf2<;0)~MuVVT`}W;3 z@|ENZcj=D_tmXaPbzWw75f~S(MXEcaUI$#Su}{@>_Q0lKkQBlMN#E6~UtX zT%PlzPV=LMyLlctkN4OtFzXB3|NL#YynJ4>i$67{YWTsUN007_S!k8s7Zc=D3Ekcs zYNrzFHdbU)$mytn#THu|RQ%nxZH3dLot>d}4#q`6JxEC2N1G)ip_R zm$i+J%`FZ|k;KgY*wA{%Cd;%#)9t@KcAXj`EB>Cmr$_cgmC^yTswby2Y)#ze2GhtR zq+`O3`fFk;lTAu)e^58U%A4!x^pp49X5jSj^*0$(TZIicu5)`h%@108Dj8bfdK{S5 z=Y?FZ3XHl0qd$NBiY)VchiuL%C}fWous`n&cP)rmoQT>k<8fGp1)^MGpUBANFvu7I%F$Y0s3T#f8#EwZS(k~+N zTpxdhVW1zoy)$wlomr>r-_g+#DuSI{l;pX|f+SscVo2XxX^;MrZ#@O|g zhK9!ZnWzX4sr~hjkNb&~V43q<)Wy>Rg$={AF{4F$qeAJ`Pn)dsa$b9A)c*R_Iy!DR zis{f_vo~18L@{H|Y-X%mQa-}_XD+vqKi9T(mP-$p?)4n`b2Yzk_Q+4kPNN;hhypY^ zwFnD{es$e8z6#kXJ=^y`uI|1oKVB-TR6Fx-x8DWTamuTo%*qn&thPgPLat`9LMG=Gzfe%CWE*~n48 zLR-(~Xyf%ie_I`-Yehx@_Y)fEzO*o<6cCKnmzj&;TjaAvusA+0J^hem*B?i!l7;i; zZLeCAg;|}9O;tO)y#1zd?$5ZWzm>ep#3zxGfb~wUlLKeEXN7a8J1%D-xaJl&1w=(h z?`d$KYN?i*V=}adJeuG02IrnBJXx92>bRImp1pETBKH=e~vQPpummhR@F( zrQ;=uo(M&88pAFY^@+-jsg@yDOSoAL3s;KQ{yVg~?~iCD>IDl)ST?@o(oG%}kIJPO zpP^H2I&Vunt?-;{Y75jA7Zw&q>bWoHIh4_>7?8n8%&O{E63U?7KWKL9y#9ur2j4lj zA+*OgPmvL*ey+cpC){JU?^At}VMe#aOYFKr!=>5!iJ6g)Z!GMyr|c?ZCG;1^N`%Rd zV%KbA)*D8s+^c(TR7gTyaf=^&W*1$lYSi)O&6^|??tRfLJ)FJy>eZ`kY$QkvY>K?} z=U3ZEEUfBXn#)@Xd6A2NXEWOu6;&~uJzg$Qo7H-2hv>PFugN=7Eb6EK+~jdGVQ?8G zRRIXWYyStVl5<_}?$``GFFb}g^$6Q+OQ3b-qobMGm;0(tF_tExL&j$?vA|?Xmq$Y1WsATV;9blqnJ>#dpo?)xt>*?49Ya zjkD0x#un6b8-3e1Zqr$yDcOAc^tRBT`5h!#_f$j}#x@5TdISq`yL@^W$#)_ydXbC| zxga90@VKegSg$=&gofaWJZ#hIZlBzX6iU0{%52>~NJ2p7c%=6D*l3lKaDa{{hUE#P zK2e{Yjg8xVLXBLN6~4xM${ySGmXk~{T`0e(WY!g+gBj+TV?h=Jv8zD2uT(7ng- zHoa!slKlabc_O@16&-JHR@aA!Sp?>KbO&e$W>_Me$jo&4kCqrNia0cw1YN+;TzpX` zZZlDNybxd{&|sH(c(tm$th*mV*y)!S)IyPzpAV)s6^J)o+?+T4R`}dlSCMA8SngBz zh7TS!M-5}+#Ai=dJvw@`qt$izwz`TWugA!5+zMonkZURvbkn9ADm=D!7Tgw;*wgU7 zIm52=zD42}%9)006aT?!}hRM2h=$H%j@&#feXO9#tGl(~Q$P zs*qnAEZRBnqQaNcvH8G#aZ46FU7 z_T8?&@nBKYKmk$5p=PI^VAHDJdObVkNXvZ{CPDgJk8`_U2L4*{91*)dJ$%D{!(17C zD*&gL)@`{3_Ohe-oBP_%A{yRYvGLG@M*E4O7HNbvw!B4VQYTj7&7QK5Ac5*s%cjfD z3r{X++E>x|-;zm)SeiEtb!>HxtQ$^kvLY$c`?O;yLf}o!SjX<-0Ljg{oOY!LYR!QX zx5gQ)_I^-cF0#tnWOF@vSKll~yZaWuc3mf3Nr*-_ZrN_dAA0Ght0oan9XEESSiBLx z%_W=pdddv>e5`=oS|d|fATA2Kp{O^)Go0k;J`2Q_$o`D6VnM-p(}hlwW%&hT66teM z*@bv?E%PkIm+o69KRDe=dD{PD*zetsZvHg zfgIB5*mEX!b*nE@1BNCJt9esRs1VFARG|SlP9? zPX*{KMaeJh)o;sjozm$$?rbi#YS_MBKgder;)iFq&EHfW=MJm0uROZVQbT{s8Ep3Q z-t(wbQgl3v)-?M{_S_S&9RFbC7uiH$Y)T^+#jEVh8!wSi0OiqM`6Wpn9$)#z=|f0< z=W1g$4_Gy)-;;B9QHal4dww`)%9q_b`#`wo;>?AnR7)}o&yiRPP|t9X?ct)?apySK zj`7|KRqHqLHwf0wv9x~WwajMv<8@Xf0+6%8#}qWCSQr)2c`LIVhwq7+s>C$$)@OZs zbhD$;B1t)7_(g@>=y=4Ej82*rXHxcpSY}Av^Vr14QHlweQRtNLJc+`q}p+9Fp2aK|#-@FLHFETD5Sz zLY^BasXMik0AK=h7w3k#e_OS|$D}0i?T6<^arV5L_H(DUEkq~eYG_(c?xN8TVd6WB z_#;jj&K0S8e7i6=>oRnyS;4k`E6jm`XZO#xo^ zGZ_sUy~(w4+Jyk++kqTaJgoqnMoUF2NZPY44$K~0B9{#4Z?no8tm>|iU&^kT1zg{! zFEnG9q)MabBLR1!2)z}Tlq=ODOeXxQ(HiZ@R{n4&9{1KX>$clg7S7n8-(ms$JRB-H zl-}9lz1!FO?3-&xBjntRWv*)!B6hJ8V6UJcGyCcBXqiM?*ZN_#Kzc+Ay>sUZ%E=qE zCxT)!GIM{G;Gu^(wn`ENs1l51)P?9TWS{b$AnH@vT#2Wj^Ypdmxa#!ydCnW9G-+yu z$Sq7%pRy-aky%@=yG~-fppHM4X5jSk>dO-z9v)+;p8^D4U=p8rE-gmRo*s)$r;942 zU2I6M@z5D2@gT+s(AXp}E|#txXX#X1AL>iQ{mqq6zwsOWL*9<`Nz}_}+Q4!Bi<= z?$cvGIziFhVvu_Rr%-$p%PU2A%$k+Y8Q8qtV$!-xZLI&#Uw{1-%aFG( zQ`z||YfDe7{;}-rhd}XxQrCER`mh2-fQv=aW+IkmB3?FswN3A2xGG)Pw!MGGmO!ew3g^5(mROBSxD#|o@F3h-+u$KHLl zpJBQEt{cFPBtvP{8q22CgwY>1*CN|YJr(o}GZ(vq3=^N_<}D0bn>C=U z?}``1#ibmFLK7p9e=?ZySbn`*M;g~US-9-0@3GO-Pf1ydGkUY2Rr$wP5uS_ADhJdGiQ!WdYaq7?rt*%Fy2;+BLKP+KE(pEF>lg9(!7}qB zZ%jPS+9RVxb|+?!6$|84V9SMCiC$`d6>5)KvD|qWxS~$6t@y>+SGu*w!AsBqN;ITO zRZMahXBnX`+l2H^E6EBPqOP~aC-%f2_fUk?{D;i}QWspz2;WH(4mPhM#Fr5D(aK=+ZA&{Zl3e z8btLMu{0GEJaLf@e^t2~-u7=O91b7OcGqd1Jo$}dv(51ghEln&EKvMf{udeDfw~>H zC0e~vn-px69yn>0(fEU1Pw72bR|kc7_O>H=2F|0pr%bPyh z)!SRIazs!qXuR!sD7Q(;q(NfKB<|>Deb_bpxnt5^~5KY z$X$#bKY~(21{nfFAe8as_1Ny?zPD7uq(~8vrFLXxt%g1}mrLHZEnD8cyR*A9cW&tB z0i%NaA6IjJ$Hw66lA8QrWR$lsDJZtLdA4Z7?nhN0ZR-RZuG8}jOI%u?(EJq{S-9y( zh>n%d&c1mEwqkyM{sPcru%C~Ir*>g+fM92Or|(UD$Ci}RvwX4U3ti?6OP|izAAehA zGgic32yX5i$x|pqM5e}(SGN)#48)-jAyc%>&ZQziMisWUEiF#nlXkQU?#A}m5&kgs z`l{uW56x?w_Euk6&2{?YEM4?Q(0U_r4coe&4NJ7R#NED!9=&F=`!clj^g_qyGkU|7 zf&|w!@$k$}^r;WYYw<3viI8!&MavKE9vc&o&nF?|@(fb2 za9T6M)(e~4an~{jj^{t^!)AMk;08W)Gj{szP4e^4stzbAtrBGS5HfV`4!AsnK%MY1 z@yI6GmsP3Sg#{cXDb5_)$1R7J?hEW_4eSLRa?YGb&JQwI*Jqwp594n2m>ax2^X$#W zl9}Y^`$%GPL+Ctk_Vw?i_Ry*?6O2rhm$1)in*zj22MyWz#9zhdMY)t9=)VjnRG~(s z7W4M@obTk0^zEg)?-3A)t(v>LUW{G5G>q%qn_d)Sv8UV_+;88$ja=l4js<6^kUV6S zc8U~A7=T*{8-=LjLVS12gO2$?I0nhDV=dm;SjS~G#~Wt4!#Togakxc|;S zS99M2VDFXmtT?JL4`_AEAjjn`qMG*EFlT3HNkfCc`Cp$tNo#cxTsYOqFF!+jY1QJg z+jRxynF-s-si(z!$M){H_tLyJw%Jy5O-gJmfcTcgwvin1Y=J1_=oF7fNj?WJe)s{E zgIZ4}N`}It+w||s0Vno$f>pXj{>H#{a7(?Rr#rd)K6_pjeOE-%>Tb`uG|#vyao#Bg z3hFaM^8P7eYMPogCh|Pu%7*=!o7e?}gnYkRw>2N(;X8@uOGmD59?wSj>9f`Jy$TSe zF{79r`{DC$Yt#tCZqw{LUQ{vJRpcM1xl}MgrcJSRO)_d>Yx-2Y68l_6!`N&Gu7tX16`magopASI@C_UN%-2H{lqpmc0o7;=d(77dQh%h?%b)p!M?%ITe&(YbKnT7_Y)!F`%&y~ zB|rqc=i}0|wBqr%+YFzSG+lh59FW@J+#A*j+!QnBs$p#8+ubPVYjMtK%2Qdh+b5OY zI|M9z6nP=I_fzB0o2OFpcR5mjAyJ4rd`7pw!|ui2iV|M?zA9c1WU2y?HJog0$liWf z%@RC3rqcmqR)uXRi{kW4r1**0U=}Td2(CCRvQz5vh4q|LFH7yQ1PM#69iv}cnQ711 zE@7p|ap~1EHnvghJyN2ci%Uq2)siJOI3Xq@n8-3m!5f+96V+sPYM5L7>4_T^a&!CSfFwVmbjkEUq0#qD0eSs4B452b z1UPZg#P-M&A?ES5{m}z;b;Wl%cpixtG?OA-&tsSl-G`HE*69 z%CZiK&$%)NySUEL~6nwb1VAxb>V_#FZm=qK17Od4E7Zb4zUWU^1M&5%D=k8_)q{OB~_bI}A3yP4ZyI zIf4H)LKQSRKVDG?22!cznjA{ebQJq>wLSeho~{=ZYLh~BNPRm#o5>UZr0gNLedTrJ zxVHenT!Pw(C=J1rzukn4oDNo^6B)=`qSa}4?o3a}G%!A?HTO@sOF0aj9!#|?eez>U zOrmdRY;*z%Av^q;0OPbE!~tD_Ju$;_7vt&L&;pqCDW^z(E*4w>T(oDao z;(Sp=rC?I_sYtKvu0J+q4P{!yFi{ho2#NryMYOAwrmVf+efM40EzK+xSvza(Ua#SQ zQUd-mVKitx32L|4C4?5o<(FR6y#JcpZRAISZk}Vj>o5HD0ysO=Auv1Z(Z1E3mruG( z6#(1ZvunP+Q%pc#NHTN5sx`Uale9ZOzc`hr8X*sP zM(y{ku=B#SBJpJ@ zD1lHRZb%T4jNEnZA5Mb{w#%HK%3I1&Y?|n=6$CDOHa!g8b)uh1|8r`#Jw*mgUfhpL z^+m;UOCxVKb;iFa_gR`+S~Mh;H&he=KUxq-w5iJ(sA&5n&x2$NP!SKL(#n7Lk_TC= z7%+p|ju3sw(A3t~j1;#rM}92Yp_*>|wE-upXyrHWNghGGj?*tB($o@I)vDgS+1e!A zx^3TmmyeV|LfQ?W+?r1>v~5F*`DFI(+4D2Mk6!A;LW z{!n^Vvb7V4Kq+WAlaZ)cQV}XC+UqbvCz{8h?z`&ez&9q1q!yyE>T32$hu|l*usYZH zfgON=mRZlpYA&bm^@W@RG~63m!@tK_cZ^WH7(lnHX*=`4+jS=jN_uqT&d!)MDXLu5 zYBL`{l<3>oydiXz}i@@VCP?6fZh6wc_7>sJyS8l zHwlGI9uz#0a^AtclHsFZAq3-%dP;-EtgN*roe}(--NwsCLG>46rJask^OKT_50iUo zlv|RaXwWcIpSPIu+NBi z7Q6z(Ex$A_-)F(gpY#>wfy>J5iw7LYI{5HA+0O=89=0WXv~NkVoWTI=v5=0PLuQi3 z9~={%$Rtc&)REPY4Y>Nc4j6jOm=Lw^^_ZdOf_?cjHo_8wr6fp+U-zhS;8ygwbF!i9 z6f`v(C?`GHS|a%^pr1fwaP`eu9pXAigC+wq@zLzb5E5#85Ng3wWWJ8uB@an89nod? zxt2qZk8{{4l+Vr%9kEZmzh4!|!U#N=nd>6foYv{wu7oz;Id$7;Kl-xjYs8iS%pb3& zySv{rLR%9#s*1KQXuq!3N6B409XI&He_l}s$bJav$F43YyKE5xpTJaa7?sB2QpY#& z@855RU{1v5Cly3+8XPr4O`>DYT^|@AsNFM>1(mfn{Fj6$APk0Wt{6~MQc^M#CJGBhI#j9H?}n04fv(!faW8NXC4J1OpifA9 z3JT4D!_N`2(*ZUte)GT|u1ia=fMqLD$O zT40H1m)c@DZ|^#C?McP;^*c^Q`Csqq?tc5*Cb@fZ+|Xy6v7}5kt|Pi=&-&dzpf)`? z5oaB8MBY4W6TrS|rrX;s+IQI}+!2{8lMM4tMGS6OG81-=>=pJHy?I%Df)1r@2T?1C zalj;3+daChGhNEub*VldiTtr@gq(@5>>M^jcd5+Mb@};Sru*J|&ki2}RaMnZ#6+j6 zB3-KtUVS-eROg!0ZA`edp@IJ*CG!*J)ULegI+x3Z%YzQ6M20{1R5)imJRIUx1cue1uMkE(}ArB2E`?xRBX1_UG9VYnumH;w#Eu3 zi%EiUg1a#vbi3|BYme~|0+q;AjE)v^_ug8U=fdnK)pRpr!1Fg7rSoyYDtlbbGhogh z1bY;-1-r0s)E)rWjh}z|E6TfDzn=j2tNtw;gz@U9%0T?k@6TqI;qR0ZSU$BeS6cIp>kK*>SsA5IAq0EKc&*DlG*M6_dYEZK{9hu$@p!f)Se8$Df(zBp>=5Og88HP^U+P5*UFZ+LeA~#p<4k(R)U|H292kd)=dp4y3>Z5ZOem zPtMqsbchQF_kwFG(U?4=NT=?AUq&SY%1o~@`0}DRo8&{AFIOwmAlKT~<<0iTMfMXj ziApf{!l&a!V7dKaMv<2sd|6H02)t}}2x95;FvHvi5AGe7Ik_z&BvVFs(+&};X#~)g zT+zJ)Ms!fZhcZyS`4bk|bV0H=RNYD+(?vtDs!CSAe~>1WJ@VRketNVhTJ162ms7eY zJ_HLbpz$;a0;t1$QJiddz^w90BG&qY;Ne4Af62Bt%#n{pmzySbJ%+GV1h}CdyGYy% zQO&dCZP)yg`}LN#i(fo_+jA9*2BQPGbl0>Y9CiU>suE??7&1u_k+lik5ZP$cVqdSR zvqbzB@U>`vi16V8*I=A7tA6_QJp?H{GnWzD9U>@+wH&2Ts5GATvOUYJ8wA!}g`OpU zo!AYoEGU8qnBlr_?ny#oCqWRA*3ycFP}*aS>uA z8QV*Q5k2gcmO85zP6MTiUIsFJMRFWgv5~|NZ6MXEo9ni0cD8k#DBb&*QFwhJ1U% zpnEvg-qk<|i^&S$dTqrT_+l<@$5M}ELT=11ZGYbN82F1t2@a-ufD|agV%6*OXc;*G z8ftBu^&77=DL=A~$l@>Gwl(D}u)n1IR56#V>*R@6r_S2~^@bW6cbhN0*$}X*o|hX; zis2Z7QGK-hZ>u+z$HXC!i^Z;{czGUBSfLxbYQxTaZ?@eXNX$OO%|^743s~0&ty^V0 zCLVLY&A+j^6HX}~QsJ3qG1F*)@x2fUZxcBOf4`I1;IR5g5<6&7w{LcGkR6f1kJz0I zFyprTzA=LGi=%!%=5AKQnGL@Zq@)HsUdq7`tKCK*-%DKU?W{4%PN3R$-yCM9X*kbyHcswiXBctx26T z7{@{)zSikl-}YFTJqk{i$UBvgGN*|k21LVjJooPAhFT&}+CN7jiOetx3MPM@aP~LQ zPtL(=#fKbl4zbMFA~AyxxNmdbT+6Tk$Ev5fbC<9_Snv>X%DdopGeDEt#5l+pq<)=d z7%ttn(*qMKM1(|QsUoXi@(uNF4GfV^FsXF}n}**2ZFm$`N7(4(GaUwZA|6>0!;lSR zHp>aiPG(L;++)1vz~xWxiSkbbtpkuxKY$4)BazzXF?WbaI*`K%^K>!ega&^4@!Yn#^SXN>;(wW>~KQ*Ap!0> z=F73;IH$eFWq6s0iw=93m}L(2+%xJR_6$U~9}K`%k_3Ev)~@c1X}64_{6*A!V)ftC zig&zQmwE}1+J{g1zCwQHFDCHjFq?TEWF~uXKpdE{d9y)6K_MS={s89=jbZ8?uODRB!P9?!19odnG zWw!iv*R<~?4m`+VA)}Gxqros)4!{MZrl-fj#x~fL=BKVID@Gh1tQ!gjMwA#9oWPW?474=>n| z`+X&q_?M&EKc3}e{y#D-|EE3AV)4VPPBy}3*PpfK@CL}Yhsap7u?cF$djpx*+ufwR zV}&0e4+RAhq6z)$V#P)|Q-!4Pu0+>yLmQS6BSY-4vSkjlPOMSo)w+{PV`F24Bjn*B zrF=!gvhOYGmNqW)>Mw>)aBV~VQ5kr}ys5PQjH9of8haBz`u|I-_J8`x|IZEM(bt$Q zUX7p#2%U*N@*TIw^jVSwT&9LNiE3?`+VHg0QY-#6+h%^1TTn~$v8NxwaL95jKudmv zK$Vip^0XheH2;H2JJohOdOzebUs5-LnPeyYB3VgTnw;P$Z+Lsd4@7uI+tf24!PWx- z1@70~3c6YU^1z1!1T&saF=vr^<#^wzE0)TsF(l5_IO#rEJZRZ?kSP6ZY;A7i%JF&` z$>F*y4{MR4NK4mp3mY2>Qxc9o&nw{Wg%6in#N5_x6S`Hlr~k_SVx$KB$AjW=f=)#K za{~L&Nu+UvsN>M4VSxoHM-1}B_f6`VJ_d?$fUR>_HV0rKAqGH&_;o(;T@>GgjdK9c z*vAlSVNLu4p^M^1BSJ7n5Vf0>3G@Bhd5TAxy}e0VBZ?GgDIbu=Qtk_&6mKJ+i(;(9 zHop@$uh~P~@!sB~I=9Q6i7gT*Yl&ii_C?08ju3o4bFU)Tj~|+9wo}TPEEY^c?DAT+ zLNUPt_A$7w_@AmqaDCx@>z&xiyS*}>HR$~*jh*Z5{Q=5yY<3xONwcvb84K7gaJde! zv9S>v?wBR^E`@S(n|O$zlRT*23n&E+npIJuANfzir}iyjL&@-OJk;*B9pdB53za;k zA;5f89v+kzXU>Nao0NAmt-?I{^McU-uy1)Q(qRE9Z;@&IQ04Ivy@|Y#lvl(-O%@g^ zT^X45IEgBq*#g34H&0bCpF|CIskb+BVC6B(7Y|>*e(eN6P^|9>2kcvLC%Qu}?O8Eh z&JVGm?gXVZG*g!y<`k67KaCVME5S?AI9jwM3q3-|Z~roLb7p_7SgXT}ubt{aC$O;* z!jAxE57&C}7*9!sSNu)FV_GNOx!0l=6)9OH2b($@f#mQ)T zNQa7AtWW4CZHwqN?O9Q{iAt=KS*ul)9d{!lgxkM*P%yvo%0=iiiKD3@M_*+bz=8^U zSmZ~m{LJqsNlyVGn^56(0+8MUeEc)V{c^n-bT`D*TQ`~MILiPK2M^}oQ zn86OJ2ghM|qBXKGWSQ({<|{{aVz`8lla1}UZFlk7q^{LobQ1Rf_fTa2dIg+Q4dN%I zzb){tUqar6mwe(ky(EA|h;xu^#0;Y{s9yp+Jb^;`_j8Hs^>ntA4XpKnJL<}ToOyWA z1@cU->0&glLcok|y)bqO38`cf_u`{AkcsI$wCX{z;;L*KeVTJQ<(rSzj|3Voo~JZ| z*WBG4H`rxSbtXDgI)0sm6kaK*4($R zI_fw|-$zP8LVXPIiAI-)nk5OpMgqa?Eb-q0L}Q;qF#&M@^Y0JcN#*r_e@i$0(pf@C zJ%_WKV4wjYol*D!-olXq)VkFMm2|-r8gD`^x9&Ur4dLrBf<&VKhFmOrveAOj%>>)8 z=a$RXIIZh%8B6TOVy#=F<>8i7E>xZG39&v0=9gfZWF8D_*poC^txa3MelSEVq|BC; z6oDV$)R?t;&6k2$SAQX^OrZU?_vsuL`xD_ZlEmapxC?;KU5g8COR~O@-fj7RnS&p! zQ`mDx2?gl7=Wxsr$i?A4QEBsP+3LIOHFd@zvEPpTB#vYhy?2>#&8B6-uN*_X(#*xP zyd3SmZ@+pG?Mo-smm5TLFE{WS7GGvvOJXIfqv@jAf1;D=|6+Rl?`_HY*Mt0*cDVhI z8%_e=D2gQYcXb!F?y}e^K3~5#1@ZMxM1-LqfU45YgBSh~>h0}44x)|mT#Q0#+n`(3 zra&}?(bVN=3H@bZ*Ep%E{r>CwVte->P?7ewE$! zBhmhc`6y!VQ7la~y+tx6T}ZNNUwg%)4eKW94B)9-u*v&^y$&zvA_2Hdob@*caJrpp z=e&c!i~pwX#{Y41|F6hI`M=aS`~QihwSkef%zp6dHUsa<$Hxm{ax(%L7kE($uKgVN zmjf{N5gDCOsT+aai4hN6%!9gAR{b5V9Af_}3tm*BLp@GUPj40>{0c;+e4w9mu$|l_ z1sbWmoInpe2AxH;y}-2>ExrJG%chGfTG5vo8EX^Q^+MI~29k)r!@cQUY}6-xw}0%r%Z589!fAY+S!5ZC~}| zMTsbtn-06PbTaR-3s^l39g?9&K73^^p|#Js6e-y1M%P`}b#E)>GD!g+!!bciB~x!x0SCw#^BKdAWtI7CqkAK(m!NEaVxU0`O z(;;GIZJjQVuo|C*5|lZ#;_l<*3kyR&QQ}rz zL3uwltIo|(J`#Q7RTgXmL~NCWP?ii5dhmUCW@aWpvdn%czZMosM~=MFvIv*wcLeP% z+gcq)+TYsddnD4Fb28d3uAo%@d@ewSB&dtlhn^eTYac&<{+x08T3ub;#pd+qe*DWH ze8*%m_1olA)6z`Qm+3CUMmgM8UFdK^d1Wg%yb|PLzyfCoXHL#nZzANMx}jl2jCxd0 z%dm$d-56wonVA{Q!oni$`6S8-6*}tAr;L>FfxnjB%2upXVbYS6R!`bUZ<05dRUd(d9N-pxQ7c ztBUxzoU21xCmvZ4vn@z%|C#aLUdSSF?t-D|0pnYOiGDy(#sA|EX8)dqGo5t_T zdNhaBOl;EfT&`cYSy)*OG1$DgE?m5L0!x){485=+-NuMpb>-o`{R{??=czPZU0vXp z>cY)GF8zR_K9_5m*C1(DH}o6v*@+Hizey%T^z&g0ihQ3dk?#~gbC-MWem;ERklVhKDOZCxoBYd_KP6>bIXiU@h;zKHxtbgJ`*R*#MozXvkT55*Znpbc{8j zdvqeNIAZ34VTsP2DYJISsYCQc|ByUc6uYi-(BaoG2EJ{0X{H|VnDw5T(BkF8#Jg~7 z`vD#z2BPIDGqFAiYrGjUn6Qp3is&AMwiBkNrkb5!-lMpD7*wDw!r0SJa_AP1qJkI4zC1x*w zY94Mm*{UUzT13`w`<2x_zpfK{2Xk-{wFu&r#@j=z4-ku?zjB4cf=VMM)a=QWq(K*V zv{~rO%b|%Z(?bTvZE6uwSN{11_;+||qy=DG&f(WwF`DoYR9Qo&mqhhDh$hjDob0)#;B{PWK zYwTtn@dnZ8Aex^2^WhV$-7V-{Hx1{{ArMzJGz3?8E{@aaNc|@8LNIii8XNC>F7|q= z(1}w+^Va9Q<}a`&o1DSnPhGcl-#eHdBVbxNMB)*>3#Wn@_xJbHscFr&N^lVe5vhB@DL5;Ouggt5OtdE*oH{x>bSi1V7Ihf-AW-l*hSF9T zy3i`JA-_~WnFwsM%GwXBW9i+!>b4-#&ckA)Z$Su+gW}>n;7?0&(1cFyo-gy1{3?VB zlWoOuxv@Kz26^yOX8BsMe%Ia|)8&1QlXr+kS4`+dtYSq0ic6yQ@q_q!l8|6&T1Kb# z4Gc8LOtA(aS~L7hDA7m?Z_feps;;c8983YF)(!4bi9V3Gv^ewry4+p(>kkrJ1qNa~ zt8GCD`O6YI68ddmkQW7W+xmG(-zU}BYhB`dP0UEB@=SoZNOR}-9X~hY5bo$n%#)yL zn7tNen@&2x@R@B5o!>=eWMm{Ar?@--W>HpFrqP*)in^8cejb zwr0&8wFt!8KVRnxNu8fYT;AP~&`J;rGWFA7)+g(gW%i&BdrW2hed!JH9Zy$tTspki z7qPh4aJDK)1ZllAOj?5I!vtTG*hAtOh-y3N#14Zq#oD{6ccdU%(Lr@3%Nkfww1J`ObqDh9OW=q7oAmeT3JLcfTLj z_|^&T#3fOhEy(*Gv|X6(+_^J$T7>+TvY^9jIwGW({sV0+32ogSLgX(GA4Ti{(4&P* zhqJb|mpMKp>O?S|YG!4X*zp4~`3He3(<=d6=cHP@2fDV12$u5}n_vK0#OettIQ8ovt7*!^9bv zzMu7b;SnG1%Jy6OBBKBPFJ!Ll}x?J8?JIfFnF za#5Ct2bzSt;ruT{5p;*R>xd638E$K85o~=QNwd<-tJQKRYrkZzDEg}`>~z9c)94V1 zyQSM^RG`u1)MgULWZ1BqDKQ1cfY-$RFrku5EJ$ z|IBnqPJeV|bnbA-hx;L%dIluZzx}$reQ;o4pfVTHAWVi93?QhktzBPlL=moD=h;$M zUvG*8;C^TWg*PV2bJT8I>k$fNwth+>r|(g~%2pDGF}=kky=9}l-?NBG&RL*StzD`^ zVz);fLZPg?YGZuY($4lRBB1zDMmtQiW_;VM zvzG3|vf-ZlP-u(82hNYQ|JkIMg(N^++cbJw&Y-2{E%i{GPgS=$`{aXb*RGM7pwWTJ z&x5u5J4gfO+^3pZTdM+wAV{P5fpM*=es}h{#-=9M>@;8uV#h_ndx^Bk&kwdOT3R7J z6OBPPK^KB5FhU@793*~JVYxmwKGi5~>UUFQM_%ViRkgMXjVyom#=uG8h^cAR)Hf8b z{3!+LX!*l0?KS|#DXf($*FAvI{ zh~&Tt31$~A{5;ky89;oN<8OsM4v>V76czjTLZ4hYJdPeeKRw$bUYN&}=rWE^?``^q zwGTe!C`PmomXF&wVw)#T1hFKd=)D1}TJ0Hi-SCXL$;~xXSlJJ{54#$Y=1e`u7JF2T zaklHp+C&yR)9#KoZc#`kdN8bHboxrCX6AzBMA!G`4D{sN0&SV1`{eJZK=5K0zxa;r zIL94Kt2oUTogx=gdS)3(6|5Av?KeLh*@cK4`S&P&hM1+$i$$@nAnJL(LxxDxvFo~X zya#<4Wh6fVXLMoFmt(8M{v%7G!MBH6gqOb{&PvBY0=3X=AwnC&63+SOC16aqr+59n zS>?k+CIFu;owO(a8{nGzBZUQE9@@h)rb2(2lA&e-V7m5`sLyb zzsDvtDEy7OuzSc#rqTV}2aHhZwG1^PV!C0gE7P+(IftYFfeGAurQ3pMXJ-#eNc7^k zhf-@C$YWJ63-2~bFysUXW9(Ar3fETkzf!%~FIyy9(E3D<@Cdc9J97qVGdVSloR$TP zS260wU~*mtNiL+1%yAH%xhltQZCCs26M(@(h_AK;E%kxK7>IVIi}yo9>;TL!rdZ)P zIZ}kEpkp!i*cT)j@Kq~S5)BGfh|_OxZeta|;MnjcQg{>pk|b7u9m1*!HnpZN>uo3f z1vX^UL_nF4K#snsm{}#(FD3*7+}@$(0AICnG9y-vFZU8}@%WWZZr_tfU>HngCRz}V z#A9(hLi_K?KjD21)`u>%v0`#nKT-SPz(=#_%shlGAIGNum56AX( zBL(*RWLdj)vZv{c<7Auzv@ydarlBMp;cQDMaX|^~H!vwB8;#quIn@P%c zGXDNlVj&)bt1a^%%gC`jR1zDc1QtO%95iF0(bsl=h+GP}n6Ij5*71ojDKXM%6=1|#bTv;DS6GiVZ_XqeUj?3y*u3w^FAnFOYWNGmD19kd zYEWjZp1DV28p5+2jkms`8Kx{@l~?c+Mbci^VS z;|v$fNaZ%j_ zXIQmG!DuI@qX%R9pqN-Us?`s<>Bk07d*XtnREX za!!l;LS0%GrYOr}9-6*&9K+h_4ZDc*SQoX3G~NAO)lCSDAn)YR=#DuI{T*oPNVBj) zk3O~N>~+pS@B%qaVo~8P=JYNUTx!;z3*ArJE?b-kvn@>US7TAC3RI&>uXUA2i`HEm*dRaC%KjfIFM2FomSE7?OixYx5YqKjn4>e+|fUpMb zOf5of*F&I8Q+5T0iB2Uh&VSxKtQphPA{bvnaW-h)r$%hbtk;99~}a0aoK z-i2&1Q1%&-E+{4~AC_(%^`#b(3K(7tQ-lN>9bk`ODSZZ6^;T1k4h}syUafC-rfs36 zjd}iaG3pVieZ>(|7mTxd=w0k>)5lnpH3qx^?fSOpj^Sqo2vm0G;&Q+@RKS*FhtS4L zD@WTB;a~qAm>nOQ*vau>#Jx6zdU789W##+Rnyc&wkz3zkGd3hl5}v0F?aWyo5wcib;CSn5b0je@^B*>1d=z? zLH>7EvEh}#oL$pO1^Ze8s<(~#&_cMI^q=`tNrg&KOxS_Xr9<8e_2t7QOdzi`;oNe1kY+F)FG|3X!2y~Z>Bo>cjRY<(BMI((df+Oa-$?OXc{uKHS zIV~-BEFg(WBh!5%^w5D#vJ)qXg}51=`=k&gwAi-6Gig@(a0Q0Lg?AnmB001%8Q_(u zV5IAdG+QM6`j>bxa8Yct6V9>ebL$o~q=8r`j3rt7G&+cP(%?jF$fVk2Z6Tc6_TJtS z5tp3FBa|~yMeIN{I`(s~-G89Tg?`rLXk#D_o(T@|(e(B8Rd5Ouc>qW?GdRYp&3+J^ zNeM!UT|V3K>^<+qfqPkyfpj;3VH@~caX8U#X=YYLAe}_+=7STi zeTc6JY9)yq%#Q~t6nAUgO$C%!zMoWwce6`a4UD$l-;_=GRGh~UK`0@dbT#0zaFzR4 zE!4n+t-ZUgKpFT=T{cBoiNl;0#gig%J{j`Oh@5XfzaLyoK8zaL2+s%&iyrwTLnW`EJJKHa##o=`{x#)+^^E%lN={`&E0G3WveSB&TjK-em@%p* z(8eTjcZMKRZ7(d=UxI(sEF|j+{$HqvL3p3G)iWozNUf4Pj|u4Kc~(+wFS_ z*F`k9C7awdECed*o!domc{BIUH)fcz%;70n(`X=Z0tyKv?G)s&kTh4VLV#ef@o*WY z)uSIZu`BopX&5-71CAq&brD{wP;T=1Hup#Fj-fmPR%q2Dlcqo>#D)vgUgX7 zA591r5Ci1+EObuJXpsjp@a^g8;TryGR>jE32)Zzx3im1DtLL0}qTNv%e`2svKKO#o zVluyO>ULk*UOQO3D3|==Hiu;L@j3I3BXw|k)^2h1@0Dx5pu0G>xsL!|qBujB$Dy{( z{%9pNhUP)~9Z~ZaUS>CUO=b|3IE2+2Y^nl?WxT#GO7fy8n_Q!EXL{$*~+#_rLwi3qbiFrR}bmbD^Zxp_pB^ zV$}q5JLL~*1JFOwf!21*uJmgCJI@%Xepb3C9s(0^4Kr5`$Grz22Ny{>90Zs#$^4q~ zU9s!el$?+5#)bb{w3L5&2+&sYFCWGKVhlD6FqSE%fN~*cW};Q86r9WWHHqw9QK7bq ze5&E0(4}&@SZds-)hk=MPXBK~{kL~m>R)u|t2q&S2Z!_jdytNEDL{9}9{O#{#08=y zDH7n4V-xmC->_jr1}b?vl^nz)B`rN?&iYqY9z=ZOWcc(yG{wy*`mtER2~iZF6X+V` zr=iS^uu39_-q7e!BC^kkvN)Ipe^Go8>9=RWZ>uCf97sT`YHrRha|H?AO}Z(?lZGK= z5JKwhUCOiZ^&Fp?zM{AkkSOzSFmP#Jw~TtI$kZQqn}(6i_2_w4 zJv}R|Ix=KTjB5mkTiTUD;IqR?n3m(SIPH?)(=fTFxLWuL-`DScr9v8A66L!{u>jx* z58DSxgR)^->yhX!_V)Ilg$=`s@n@8jj@o%Hx)b+92ax=iWxXfOm+V&yANQOc&XCyf zW4rqf+=~c=n3peK!fKGf2pQRCZfU7ZCv-)~R09gL@fcNi^N{)wR5`Gu7*wpwPKJ_} zwR{*T?FHNdz|CH#YU=A9-Ivl$O18GGTRxHux-gzG#RDwoS=n}Ym3hSOod!?wTHcwZ zT~orcf1%}nnJS144Vkw)uhS0{-xPd1x?|kTdlkd&ny(nlX&aW0NIy&CQ|s_XG~%p^1y} z-pCN?>Y&!GX zPoJUbs-Z5)*{f((Jdc%;0rwpw)+r*3G@!s11>Z;%sPI-6Q*FQ&-tN9-Y^CsB=;4%! zyZlbeK5bAP`N{HA-{$Nu0QA4dH@1|N#+Xmt{gk?#Yu8=tGJDTd;NS3IeaR_)Wq{_uFJIh|%aic0 z27USRJ9R}{;jgd!<1w!Uzq|=&r##kZM<3>wuTeFD5ud-_rX6#NHz-vd_mr=UN`o|1QE=cL z2h338whFi>!SZlfrHXLbWu%kwH3-RUjyvAo({zBHOY_q8 z-#6_2U-Y1V8jA#HbN=rAZm=DPWa0K!dF>Q4KMBHL_&C}pb9n>>Z{moP<(EJ?s{KFB zeR(w2ZM*gzrIJ!4l%Xh+G0H4blp%@=q0Cfs#L+pMT!oC-J*~*L9uOc^=1c9%r&1??3;w7j>SEogKs5 zhg$42Gw&1id-d-Bw^e8~Qv5%Bf&X*8j%H?NqG^9C+V`%RPAH*Y8a*BI9$%wOCr;O* zY+r`PK~d4#OtJ0TBhu5Qa$GuxMgz=Sf>Vt{b;#ym2S%Ef!D`(J&=)>z7i55lFsi}+wNM_ z^$cJ)@q37fh&c4U%JUYk(AC`BH3-f=AjM(WCKSAicl68lb31RfYx%H5)oKfv4rgUi z-*@CIiQjts_%R#W?DAZDlzrpBe|N-d;F;^cGCApuKV1n68*}(xKWevO8?#HtdReMx!QW& zRyvj#@?jp26eMIbAd=CSdkO8@XSyqvV*nG(8CQdE@(G!pE{+`4*Q&3DXH*}+=SzS6 zwXSxgbWY)uud17y8~Hw%4d?i#7WIvf8=9TM@Du_D0n^LiZ{@ui83~=yG7t$#kW-_9 z{`RFyi{a?%Bp@i*-XU>L<-fi>8S2Y)53geq;`L)Y*I!#T@4i=yrqSWJN8>YXdfwo; z!m}$;^5~Sdhe()j*!HtHJQiZ+K?K3t6%#^yJ%<6@>dNC7|uhQm``*;r&V{`AY zEA~$9`5fne{5YfI%2{e5_4bhtr;bYpB#4|-#`x&g7)zxL@CQcw=K+YBa454dF~XC6M5)H`kv0b(CZ%))&Yw&Cn*Gwgfe(ir zo$HB@@>Z-~y}A^VKguMudTpM{ebSu|%!DdyjNG5f{Jb39rmNzuF?42RWCV(W;PdAx z5FQ<1U}TKlZmpTYtGa1Z5Kx}&a0qwwsW)7h5&-&S`*}5FW@n6 z+_<5Zed>cZEKp8D>_Gb{BqStl{hH-fc{!1 z?3YrYf7KwPi0IaQmmZ=aG(-#*h-9+!>uB(%#CU}DpFe*Fad`zEhsM4kYqz*4#ceSA ztedbSkf|Qq8-bs=QyqM)%MC98r(Z2Upk2#!?a>^^xGETU!y@-F+(#VS^P|6faX^Q) zHu^`}ebYMXYcFf*>)XSr8=5^X)shRd8a%WCdd^zxSs3VtT)$2)4Pqp?JsOo`(KCH0 z%}gQKDt~IW8IyR>#k`~oc-tk6AYx-vhna(;M0jXuCN*)!HNTN=eo_g5q9r~EkFxjM zq3pd{k$3MNz*}_M)g6E=iE`%m&y%^lI@G!;Gx{}UrsCMz&_-E_1;pdb* z-U~4b8A^g;;xYhzC5}Xyk5xl&lCRSn(VedTQRJ3lz2SscTAufa0DL+uVBxiR>C&dB zhQf#&))QgeMpkGjWJ5@@#8x-UI}?Odu950)pWxlSJa|qM=JVRP8U_XiFn1^jOZEO# zk&MFqyi;o`H5T@9P*4!uaHF7DYs3#tmQ)UXqnx9suFimMB#oPIB^u`Msb6OP-cb~X za~M~yS@HTcnWypt#4EXoA;qkMdbl~uUO%a7d*H>^XGcV@Y7!_vbLux*fNn-a>~?T? z(x+M4A+MJ9+R7zf(6474!0Zj|+D!Q5-~!^q$Px`~3^L!1Tjnp=wzae&*Oqm3IN}H= zWpP*#ZP~iDdYj$8;zjeoHAxhPyi`C=?iMaf6CrbMjA{&8&f3`2#ElCZ9V$$t0p!#X zcOfuGm!OBZ)u2@k*IU_><(N9#6Laj>hhC0?)^mXcfiq645#=$0?_kg*~CHd zOmFp?xT!YY{;#w`e^L7Jew0_D>GiSk&bxw#OXd@xU%rN$1I7!XH8ui5LVuwY1EaNM zVdFgkuZ1xTOryRLIMOjt(^m|+`Ayf9)V0VCYqtg!b`~Wj@`HnVx~Km*8yecKVPb*- z>vi%)An>GLUfJb<0OIJjjJEZPcnsR^`NT;yQ7VcTVlJxy^7_*yJ|!-@@jGBokLVaM2KUN zsZif(sAiGCSfcmkQgE=T@9);g=;%uL{b(FJwjMnSVh7tiWcRH#62G7KWa*1)-Zb#* zut9j!YME|T$qZ-Y29y)_B2PeINF!vTG-|*ewecAQ@(m>fUs+jMA=u%=PYsq-7pz{d z8lZd&V&}e*3l}d!)3RJkOUn!+1tA(~JykTt_!I~B@=z6ThI}`~hW_BJ-@iDKh;Y1S z^=bXRt_9;+g{0_-C#3*7&T)-4GH41K@H@**hWLrpi zV&p+wph%QM%JIY3*eHm~lncRA3CKhMF1$FHbzZBRYN#vG8r0xe~c~bPk=umV~_UHYBJSFdRv4 zB|}Mx56G9+TuJ;BTd&ji$z^EVI^@=T1I9H{Z??^ zmw>Fq5+fj-TVA-ux>M`od|3SDZUv16j)BCuhjqXl%i6DA>2|dB@VqHQX3AS{pLU$T zKq+xujLr&EBd)$i<%HrA_oG%;0!c|pIB@F&0|QfXK-dqbpZTn=!;@B#5D{U?opKJ4 z3vwu~>gf$rFhC=>EgGh`fal5PPJHKJ)A&&9;p4}IZ)BJl(5P8cX&&}t?XND52(}~$ zDp`2s-=4ZRS$|iF9mcyjRnEVhI$(FPs@Htu9rkT%`}Z#cSq#?q?!&_$#uPl+MFQ>e;j^no7}->6M@`Jn;JHlM3%upWB303`yxXAa{`9hwvGWMf6qSP`IjhaX#057-Qi<6w8spz7I%!2xK@soo|4A^zq z-1}VPQ)JqB^))q_u;J;D_hkB%4`}PLu3ANhyIiKS;z=nkZ>V%I-D>C4r%$m5l|1Sb zx^?5c9u4-|4-KCB=F-54BiZ!ggsM)NsI+uuwL+aPwT^1$snok&zv66)8B^LUtpsMX zO9GY>YjSi7Y}&aq8qk;Ay>sXV6`r-$i&o_W;9>dlO5u(|a5^KNll; zeNdX zpHd3%oI3bhiPx@VVnpg;g?%Sm<-L~LURhC*4yfnZ;is<z=U=NKNfn zR^RJnDEl<>CUF)>+rIgMEpGsQ3@+$5==ziQ z8-H#(o@z$-mMAbEy)A5&e?)?@e~zB()XJoWyVP+g1qlvim8#qD*IyZuy@ohu8bu{C ztl#X16SKZ&-$m}S=m#z?PLXlcWu@ayoR74l1r!zUAS`eg--?VZhjnH(Om{JgNQq2L zqD_X~6f8Ux_G9tNcN+z^sL#2SjMl2|;_OxN1;Vv*n^BtM7-KbZZdmp(6uCJqXJL5= z9oqNKlEkZ5S!yR^kRUSU_3dRHx?0JNq7MHNJ~sN=+RSK@Iu|l|^pnGmW3TOx6?$jT zLL~54ti9L5ZDh}WISB2*(!XDvXh0SOBfNExRrWwRmht=X@m@XDi<6KI(BKC6Jg~6XIRRZrkGbumyn=9zo4!&O=EiY`*b9JtQ2u%ZxQ=(mS3N+aV0A zb@*jkSFIw__BDKb^-wv+Z>#%se_>U4M{g%DtV&huYt*VzvARfpKMeT`(MO(!!)ElL zKf>U&`mQ4q;?>d91$dN}!lfS3Q(D1sv8`!ECF1+6F_TwlBqhlP!sI;+i%DQMeYm%VMk5Y@NbwBdc7y6@R)BL)-Fi7^T{kyr zMA1jkO{{XVPN=9n%DU2iOjiCD()3daFOR6nsa`i~?)8-GpYK*S+UG66H@0}ocn7DE zTYgaoxl}R~ALlB3B0001Oy}Xp$;nFWRw331Vt~>o(NM4lS=98za6gnSqK&(vYk3-M zbNBEu4);o*DRx>b%MzlHsPKC-C*@;R@eO^QQxYnHbA>eF$Li|pS3*Pc`^M9@zL~vS zP}R4Rh26LH~!AymX(#Y zLQF?jm$Y9_PER*I*vASxy;o2P=paBWhe@)say(!c8l5GxvU29Y$)^U7jI->n8S-~< zY&+bAikHk~#;Wko04(-Lyo}#va^rxtV2t)F_&LIRbPd$mS!Ra| z_Taupf+mjhSKPc>*_2hmplYO3&{vFBj6}2OBom)9SQWf8@kK9%*VD{s` zOB)z}&ACHzB6|i|WJXn7;qZtX2X`_0{5pFH_o%6aQf2=u z;&gw{lnXWiX=ePLdQ&g++@IFiWGcK25-sPIzyGkkGdA&TUMzL6cb=iVrJ1IZo|T!2 zv!OS;QttHMWTqRg%M#craY6ii0QO)0pJw~U#&8*%+OZ=F7j9&1?8(@WY|J^}3#*)2WS?^5v~(=<)kF5uv0~whu#qd3jyW7a z-)B(hBWRp|_9W1US}b**Tc37PXHKg%U)EFLAY_mzK}m>{DLfG5{n>@P!%*+vma=3J z5fOQL`r~!Th-*{B+ZO>hT8w;!jh$UfN9TdTBkjw@yKpPw0Z0z%{#|JwRZy-iRCv_p z%o!O@(;F+Soj#@r}^ox}Z`KpY)$m#!cu}yEeFkk+HEJTyMF4`WE%}*T)qb8e+fo zw2CCI^O8t4c*G2p7!u(!aJTI21q5V<8Z(Ix(yjd+x z%1Bv}9%EAA-jw8z>U;K#OiUyL5rm^)1j^@8?8mnyfq}KBLq%Pv1?xUL-^te8>wlfv zUj+xzgBI-=&xviS8viNLr}P&}rX0r>_0B}m8VY4yf!5yLhmkh%@=bq;Hm>8B&#_xR zi|=Lr`Mn5-sn1nnjBCFff0e)M*TIBc`pP4VDVUs_OP#}j3`=xQ#*_`fc)cE}1Zk8E z`naBj2?mC%x_*5Y-m6?=BaOjLhYbu2zWklK0qtDwey*IOzPz{jRpJ55CG{BD+1i@e zS@7-h&!ep{wwI6F5Z@P~JOEE0K$_JwTDA`(D#pN1JWUNv{a?krstn=EQ#yWWXBzz{ z()LFBuq|jVwRgVe>Yv%)dtXR#5-!0F#+%-rpd$cxJ94S1t6!k~f~Vt>(7A%eebV-B zErU0!l-C$qQtM2Rc{D0ibTr3dRybpNZ`rH0eZQL{r(UMV4}|#pD)l$s^bW)=E*n=R zo!gFTZt)ov&#_x;magL2nodL6(UM`k6q_){a;GCfIe);C7Vk1fAk)OPPu%OHK${jw zh7koIY~>4Bn`wghex)RM=3iIEeD z<=8AFlnG=k<)giw9Zm#eDGrYr%eye|DsVqlBr5&=oEva<{yzTTa{y?j&15pBiuZdHIWN{qm7Li=5J>2elbfqS{R>ySuaNfIvXua5@KU(Q1?4J_YAmu z#1!-pQ94E5_oqXU2NsDy*GN$7+yot@W19NUd$eU_4)5rZS}Qc8yx3jDIXPzP zq}jDu=#qhilKLh^@+i;`xP)$^ph-yLr{W~il}!IKIs1B^6j>uPpW629WnPV|R|<5$ zM9zE{wTU$Rp4fTqvL|FCfJ{gfL=MRY>0`5Mx* zu#!+4~qh7~~?#D%O5|TFX0x=&qurL9(L&({TSn(JPC;|4O zqSopq?s95DY*sO&x;=WkoKJZWIzv0mC_#HfG*%AAhbsZZR3l|{q)x3@@_hcGCkht- zvOpZVaoGuO-h2tZx>ra}n0p_t*JtwfWnXzg5($6;6duL1)o!k*YLQWo{#Z8P!~5_eH2%)?L57 z1kt^%j!z~6TMulaEJ6#oG3f!%q$d-MF#(I-g4DI|s41t!>fq$I)zP zWo0!9=Rb^*7OAPJA=mEVXNN=3jg!~Lgt`DOm6E8Vt^zGr^!O^}*1Nx;w!J1gCo*wr zR$7(Kv=MiD7|QTyG&;%PK`@+UH&Wy}r1&+})f@2=P#mpukW4#nxb2ogLalSKGCkzB zb98O|k5fgV($1oFXn^(SbO!6eFFG?=mil|$h_hSm#te;%V3;VKdOD{5mC>XVCO zbdXYN!Y01`7XZ47ii)V}=>ja8N%mbdt-H_7_>iInb%WRo_nqTiT^7I?Uc<7!9LYLt zNjIVF?*p(2Y7EZ3I<=A6mr2hE*|q?+@VbM@$Vs)4KLG|nV8uzz2WvpneE3Yk%-jRm zCP>Q#5fBjSaao|BELO)_2lKXo!|ZnTEL1YG;#Tq9f!Se9xMq4#TM*b603j%VQ|_ae zVdg61Gin1c@-;S*xQZpA24IXPHiDdTLN9?jo$c6i6FF5=uFHvyE>@8Mr<#iVyvPER zEcX}jip=z%as$k{2UlSW%+0I9`IF%S4+J6%tfY6(j=iCkh*!^rtfl&mOy`{gX5M`; zb3{ss)}={YR@`87z-ib3#6lz^2`%g?Nb8a}y4~A9#WlCIb%|kb!Dm$plnRdB9%Q-L z9SD8F)z$m)iW%0%Q#<~-HpVapFETFf^Z<8sqNqX6UD1?fAuZD)gk%#WT zem5H~Tsi~(Ur&R2I!&En&xw9HU4Vv%3-~Z_iz&yVh#Yyg7HVoRj+v#mscEl)db)!k z`|=@$6KJ#z;sA*YC;+!(^@8290W~7`S<;2dGWV2AWpr+vdDt1wi#4r`vJ0g3YyW>S-Ewyj(K53lY>4j>8*d%W?;kb`_^y{J?e8CLkSMo zcp>7DiziFMK5*cBELrDwo(5E{uf4ThtuKRG~f ziu%bDa%)4>4C+Yii&kg1U5~0_PBuH5DRL`RUhj{xU;~&X=s>pPViPBX7p^gY?SJ8& zj$wy4!4y>~-~J#B#(+#M1$?eNH`&v%DtWK#qrrF#6TorFz#$r5ES|fiFT>POLX;Y% zqrsD`_4mILgS~tGqY_lX;r;pge0RdpE|V-q%F$QgJ4$hYCaF%=$)^9^Cok~ z88cu{y95(4h1_ixCyaV~2XAcV9~*vcBkiLE)@DtF5-F7N z)h*-Z?vB&5iUOFC0%O*exJ|ImzH-Go`rmJ0V`Iw$IfCC$2I*2ZZ`tAs|7^*9LJR+% z_mA$P0R$o|)1?D!}y5-po0|Pehj0p>E+qm&Of`-+O_28c#11Mg+ zWXT~2*1#pu(9WQesZt8GS!{X|Mn4L^KeysLNI!Tt?yP?3NV9xrF+a&T-C_|Os_z2} zAjIC=w+9J$j=$jXZMY7e7qJ%pvn$Z&$BKcT?^BGz5i`>RW__<}S|_!hDE5!V_co$v z#StQ~BYu_@E^IR(|1nhnD*TaOk~oN~lpM150=X0ZBTE2uuDZJ*{#aXY(fBiG;YBjy za^8!`2{)Qe0s^lx3jy38S2Lc5oP>$^^*k-XF=6xC(uoIUm&^U;c=l+8SK8V^*X>#kYp{aKR7~fL5 zZx&{sokyaunuo{Cq8&E|VQ|~Izz%5b-+vESMU3rKXEIzCGAzrP&`99mK8?DIZNml~ zEYFH-hsYsdPUc(KPX^tyP|X8KW&8mo`dF$TjQ)ZOJDvrz@)!3NlWPbF1-PRf=r8_^ zg2E292g}~b#z!E3DSRYY2BYkMMW2Hev!lq+DjxY8&CSNBdgL*O0;9R<5ce^5q5lDG z2Tx1KNjM;&E!N{O(nc($+vE>R=o3P&UVVs6&c?>(ZNA58i4(<+GJio@l7X;>W~5TU z*9Fa>#joRFTFzSB%~7luDP_?Vi;tn4SP7d=>yqkKX7hq;% zT+hH=a0*BU&ZZjDi4|+T7D0e=lAFyEmDZBb0t&J;v_IFo#S>&NRsX<&M<^G`(!eSs z(bL)+rikYY?>p-0UAOh|_b+^8G=zJK89#>bjd+a&oErZf1}LJQzM5@rZVvU(YqryK zmfOg}RQUW8tM-TFV023DrQ^^R$(|TP1Cw#U9Cq!B_D&RaC=+Bq9(i&6F0wJM>J5)e z*;^T7w7Q+`61Jw{5?_ET$B#TaoCF*|t!(TZ47r-~JwyqO|KwmU&ipT8s{~Co1M=lX&hn3FkZvk#;;Vmu#5(EA)^}L zIc626QtNOPO&m7Vd9oi*P~9#go6P_V014U@@{qwH(ZHaOJbO^9eJ>8a{>11-Lc0fI zyll8Q2}kur<9;LE z8kD#_`{e}#*DU;6<$eF$+#(ru<*dL3-El7CE+5LOtJ5VgDt13cJXqrQ$e!>@*|Z4W z36v*CQ}67{S?V}`bte}KBrrEo4kNE+CJz!NVgBhOmCrz9znWVb}%xNB&)dhe)6Z+)N?yFF1!AAWdKY?b}>& zFVMXl(msd|N}!JnU1<9IQ9XWnyT0x^u(U8<94tmx+BF13Rq<8=@TDoz^j`z;t z_K`IlwLdMweC)3f6N78D{T!>^`M5^$-3PR@$R%K}y0sNHe)EpePBFX4SN*b^1a}&hN zJGeJuQsRNNsi=)_!U{czfKYP#ooP_K5TZF3N}WDVu5Ox0`cM)GNVl73v)K`m);<}So(ID^6p3N7^U?c28>Rj~;YaAEb)I6ZcdiCk>v`DZiyRuQHS;Nazi zeqh7IBq^v5EEBlAHfhZl;)0r2yFi5A2koWSQQ=6xpV~qc3i4+*YCd>J-}Uw53w;DG zs9JKwYUhD)WL}2ot&yHXRB>21iWw51uC6Zi`omg?Ci$uf!e=pVmM1`jKE4e{WZCz0Ha ziOB#cPX5_}SNI$Nm9tFRmi(BQ=toJ^1L5{P+!ILP_>kjZR74pxmH@tpiPn9T`!O9Q zQ&l}XQ5*B)rbecG0XqOdv>Mf|k7Rd51B2K2EN?+v+>!Lpr*La^e`)_TIfs*^^dp2a zCx2M@`oGTZZGq_vxhzF~_wQG|A?yD);^+UV-Tn6||5vT`lli4v{Fprln4i*!qFzfw LZ*Q8q#fAR>NbM2C literal 19434 zcmeHv3p`ZcySM5mmHa9dBBVs+mU5R%RFu#SxeP@OxyX;#KJRDuM~jCC%A=*l zqXk~$d7!w%hYy3dR&J~uP%S7};K5ZYeIQes;(UI;2$&|yyBqo6etb{#er82NMeP+$`K2T1q%iiTMfg2r9T#= z4g<^A+p)2+7)>Iu^WN4Hizk7EipERy>p^9tvJpsISnD$AN$GIPgk|$8q8*lz0jU z12e_IV3=6g3IR+trn(=~&%r@?a8NB=JXoNA*T8`X1xEo32gl)5$C_5h!m2S~iLC~! zRfBB%tNS_C91bO3ixLl|;J^}30qcNFR_cKV2dqlraQZQ@eoSnCHCQnH)nGl$O4WW& zKZi3mHU{zrse@EF6dVVv(~sfwS98F}U)h4>K>{2O2VSNU3TkHkbz^I99v<=Vm2*|x z0OKkT&n}*m$Bte=KAcEf4(V$1PjlXT%Kq#V*FUemk>5k|Qn2a@tj>BUlapnc+xlD; zqhdpj^zff3c&v8r0LC!~cQ*NHb$lHQul=QX&85F~Y^0{3nt-7rdhsFFJRwYp&(`%s zUcWICogACog`;cJXW`koVXxj}Us*M5g#T}7W z>!fz?VQi2(cIAKha*8ocCsp3tX?}R2|Ff)H(#}IomvoCXI9=JC*2W}}qBm6uH3JM2 zdW0zsT9FjIj2)6m2r(&pRqD$g!OTEUILUrY~4{H?wS;V3v0_mp>MyhqH!Y`Naas*QKcD4h`y#c-nIR>n?sp z-c)LER^ZqNqq!Xat8c1y24Wu+DN;3+{DLZ3;nA9|W3S54Q>9BesBrRR$Hn=yd2Ivt zEi#NGv1VO7nI`yWg|>R*`$?iLMKYbW`&Fi6Gu8LxAX`4TMG!tzGg4ZX%$}knn7ijtc(w&jr;W`x3@OnNV!r;k=}* zCb8bnb4t)5y&V67`2HFCRrGrQo?}#zACXfz$7o*p!^yL&j<99RGaXbmUaHrQkyG>-Imh# zm0G0Y5u|R%bfpw^IHa32w%5PW$|@Im5}252&06Kiv|nyiN=kND`9#CM%n@`ndKZB& zHbeV+?p^fqPOpg>!{u3ef1=sC~kBo}yfg zd(vG+u}3-M;bv_ja37#&_@>61g$=vyZtkW^@?9o)?u_luP#H|I+c>i=yxNhc*o(BxJOB)ybvb~%#hlI=RCE?3BSk>ksEXce0I8V(_i zN87NC!jW!75~jLp+iS2)|#F*r^xO!v2*flmxDhwOOBI3dKkf{GHtr$I}# z_R>chGlZ$S#MGAOd)GZ@hw6zf__8m1@aoVYrK1Dr%YtU*Vx_!9K9#a%^=q+xIv2G< zA%s1<&#=8TkiAXG`MEg#5F|18#um6|M4EzGXYEYdfd5Nu%J}C4v2Wd9pJ7{09DAOr zlqXpEggyln2YMI3!YZ5G6Dv7_dV647UcfimSCC z-4i`*#eQi8xLI~T8D8bj&RmEZZJ-9X6$cL4*e00td5w8Y!!Ay?pkj`v+jONrHgR*z zeiVTZ#QVKYQ;5Xd+SYFNfuG7u!Xe!b<#;&Yk?pUFiR9M2Qqjg9R0S$?-m_@i9sk#MG{7H~{Wz5Mf$$=I?HliT z8z9;uwD(H1>?3)?F&y-iG_{F}&h0EP0HSmBY?Y`3=f+gpA(_f82dZ;)77s9ThvhPS zBM_vQ+SZdiFM@1+1lDVhZKF?r2~X_dcfFFVRDO2=vejf+)xcfcnbrV%v~F}_2Pys- z5WOmwA`C}$tBTST)cyVEb0QIQFH82fSxOq8@$jkAph_?l8^}@hnOR;OqiMFAlmf$jRCD;CYdXeZeT~@C+`h_Kq|h*-^$^D ztrYdW$*jGdwgTVV#)e|&mSic=oY zv~o*Nq(O8Zx`XDLFZ4q9*^*_qGAP@&%9Kx&`%r!iGgHuE#GeP3<=MAGH;(D07A(Rz zx#xko18zCz)ZHyBvEo>@dudF>!;zK;9Xcp^DiOjP=c<0@f{{fP+8hH9zqptB*yPA) z(TyiKWoGTza)WHsQ;czp1FhF4t)uhm?)ci?DKQeKOm5 zzmzV~Ml6AnR7ZnPO~ick+g=EHijM9KjnnFx92Nia^o{D{=EgWNxO%5t&svDh))MCo zVu?tp&q$AzoxYmJBy)DLdSW)%k zo5?7^7FS^BncJa?%1pwZJ0HF8=6i!|pSjz$VoXsQ1V7s7VeiA|!%pm7vi`9Bvb&!N zAgl$kmh7_3$b|KqG=zssm?xp4LpyX)3RK$2XY$tKxRC1vwslY#b7?w+BbrRnZ8-5z z2NL{><;wWe_6d%aLyqQ4mAsN#cySuLv}!;yN8GIK{HQyrxGu|!b5(p>Vfa&}2I{DC z1MJ2ZXlYVg$VW_eep9qtB%=GJTOcGCD|DVg;?Tr$`Wrc;B=@$EA-6dHylCdgLMF^c zV5Tpc`VW}3 zf1BcOJqd8)_=|5gpWSt7ig!2T#9z6We73C<3A|wpY%DM9h1jb9Ca+hkRV?n)o8|?P(&Lf58#es0k)S_LE458; zrl#1ZHc0Fo*c#VUE8le=e7f=L(|TH3#uTCSe(gjmCm5?L#n>S7RVnw0i(hgAgeOm4 z>n%y~76Wt`rR;1zBYU0HE!z{3%w@+5 z<)$m4otm>ZgW9n6E?N~eJ?4JFu%S>1-F4qZ@!rA_y+b~{nSI*v!7{W<>#vE?6cp3a zPUq$5^d%yBR}sb6CFx&*K zD3I@Nr;*A|4n`eCQ(_Q2qeXE07-^FqAo3HfTcsjpDASqs&x{kXs-@K>l zdzT~CSN~@FP0hcB{0}$zZBioQza0n!x8HEFGW`4-F!|mIz#sy02OMPFl@F&eCRcL} zAit62H?n-4sr(+A|1m52jV%90WcfYY`_DGF#q9+H0;{rGha>(R7cKP6yt}^vZTqet zp~qf^Zjq`&1sEM&t2B{RGBpNP^M0RA;H0VAt>=JmVVH@4Iv7gaxcNO&plqRsWS#?~ zr|DKo6&%8P^flP}KF{xrLS}S5U|(ylLOBO01P?XNLft^V2Y>M0&PYFGY(iI*D-p=O_gNAf_1`%Nvn7u8*{O_xbxsYYptZinuvxH^v>hb>^>h zpRiM+>EqA+vIel>b9{GoKDE|(fti8tvCUJDXvF-~a*=C~yt(H}bjlTSMIsLW4!M5M zP`;%K{&N6q=V@Rlm^0&G42T(WMe_pN>C2ocqWk=CV$Bt=URgnU7fy#N&`wi&eRTv=Q$Wl{83ziHXU5XwdSpr^-A)5KSCu`jrR1bM@Z0jHJJ-FN(3 z8;Kuw)o3U|lB!BHxl#`xnZ{VzCTRJNLw&9n$_Kd{Q(l2e+VnZC>wa39_WnlHUa9~$ z#jr?(-vfUpd1*D}y#N*gq0ftRS6)NZF5zU-0d)?&nGRW~o`kfQp!v8tpJBj~6(%k( za?9=)o3N3k<^`A34++wt+_c>)7KazE?C92iBgIAVm*ma07-SQNJ*26ZcKT1k%#a?3 z6_=5^6JYaq7vXkrRcjGtBptEo%AJptxJ6LresGW>(GJ_H+mLZqh5_SF1gRJU12ahe zN%2uBO)&#*s}Y3vos18V@2});?iwOR1I(PM-{JyhQEBSh1YSvr)_o!Q);> zxGhkh0i~iBLfZ+;U4Hc?zj6wTWKw(kQ{twrm^~@(y+s1Ho#peqKTADMw&?-_<1s)6 z{^?;nQp{Yo1o7oIYzK8CcP;QiQ(t#wxj+72Pmf-UM>h4ObgkR!ycJ~?vf0t<+w|qM zaJwHjAc1(u{8LsqD5Y#+e7k@wMTDEzePpy)yP`etHp{mCSNm(xXEf?+%F{)twLLtY zEfP2-ZdWlRYHM?5DKm$u!cr<8HTUMJ1nko*1(8p%mYfxBUAh?v9n+{x;%(CxNdTs7L)z3u0AG|42G zWDMmI*w+ueiIb)LD!TFtZM_B%7D2h+*U798`Y;xNr*lsmMfXu>GR$6983Z-`Gj8iM z!m13|DH8leKMf?87_h`&bx;tAY|=|c8Xe`vNFjQ5o|%mK@@O-?ln_G*`?ZqF8DK~f zD%>_mE7dK!L_mLF!g_wPr!xS!nJhHe^UJ5aMh=8^|4<}i4Y%yT5jOQSb)RI*RLx`9 zqmEx>6DONo-s)-QTQ~KLTf8AFqwReB!*f7@apO<$9a3St55FxB>@T-Qe+9=CU#L-h zo>-|v`SALm`(z9dc?%7Ig;clHpQagG>80g0(GLBzQce`RcIqd_D{2}@JFO@o0Zer* zaq6J-V(Nr8H)gDjLd+w=No)M^pZ|s7wUNLk6N`K^2(!&Xz(1*2o|=>YE28{@T_F{V z6k6FL-5&TN?yKtlCl?YH*aJg!vnX4VnXjjZG&g41FsNm^DfbYsOl8d7G{_>{n(?LS z1UnxpnzComUc^DM=BETq05NX=%-Kq}o)eZGRGQ{{Bnj`khxpZj!52J7DCBGY&rrCS zvgLn5-?j+mQrdL$#u?#II9(j}IPda5QP@Jy9Ls_ZDJiP=+(GjFq>b(L(owkGDd1-v zZKq$GQ!)Nk04IRQam=^HPoG8qtkRq-lrDI zH5P#ro$-wbFUIwA=~F?X_9uaI?qRkqO1;Dx*%(SwDzVCwrkZk_KwA_dC~0w&7@arN z-g<~(f5^c)0eRf2}D7fgHlzD;{^B4W* zc{*EV-{ynzMls+*X%BBUb@>{&5PmUM6kN|F@SBw`wm#B-_j;!)D5^l=hXKrNaAEqg zrYk7OFNzv-ljb3-SCxBx0(TYs+ZjW}Xv@=a*{yPLs-jFf5Ui5s2hIU?PyXNqF7xJP zYg*}6%SerdxR+TX*zQG)pSh`_ljh=LWA9UeF&9M0vGfB3Ses9lFRZDN7#+Ey z$z2u7pr)x+XrsQcmq@7t`(aK7w7HW%3{g26u`Nk~YRKL6{}T)Pe`dP>e=?po2VU5w zu3*CIRZh1`R}1D1gI@et|DDL?yuT{s0ziicA>c%Z+%86x1c(hA{dV&0jp?4Vs)BQ^n05#K@YMp_i}O2FLwcyG=K35>FrK`i_>>v^XL?sVO!*N$Fl zOWgF!18_+=XNmKyq9phStafA}&B2vmPLBewJY#bE4}|sGN0r6bDu(eHS;*4WPsw9U zxiPy&A_Cl#8~A1q1IKsIy0MwLl0N^sAzRQ0m|$Fafr;G+=3hWAUsi*m+Dg*o>n>rC zs(6fR$4c_>>)z~YMWSs?Pxqt-RcB?r;p>j!@pC{)k+;XDm7K!Yz1$_+=n1y1$#~AI zu%F(-hL<3<$Pz0#&u_#JP|4-YAu1Y+CYZF>f@z*i{db@Fui$ADdwfZEe$e|C8vl3O z8*VJhxXXxvKc%j2*1z%BQOSsk68>R?oC+xOS#axo$zMTu1JH>610MXlevNwr+`?cR zK!4wCB8g}WL}+EOz#R*H4GTWHh8!c3TSK!hQZTiYm=lnuzB_4@d3-ay+ueK@7@bF>*|mHodj4J*tzFce*Joql zSw|Jn2;$?n*N@$SG3cXx(-lc0X-_WO+uPSr3p(Q!DE67B090i{!bl|zLcGKPx5zKL zvLJ}AG<#OB`}|a!&x^_`z55w0%jNF#alzgy;@7Dvgv77{pZ50j*tlG6M^!niw@G|P z{+Q)UtdRL^Qrqi+6Og^sfed5dMGc`jOOW0Wizrpr)2l-;s^AI}qq0YLLT*}rcoILS zfX?o_4fJJ(Ydg+~wtJp)_o{-u)yu(!DX71!D6y(9YesVrrL?rF8E~h4aF$#^YY;+3 z&z~P6F!1+La3^#`X2`B^+d@-=C-2IN;m)=-R zoc{AKPfLZ{0`BWlvNzoW+@Ka|~_pfVo$N?5W!i$ILyl%_-$=ohsgSDiinMrRDWFWZBa4 z$8h`9B8}~fmV`G%XQ3RsJc2oYNIcSe5?D%3E6X{+V84IjHY^>?l1)U`WlbM0-u<|9 zqW~i*?IC&0f~HXNBp1-J^)VrDf`}wlP&6_$_Q&FFyxUXvWV>Y9I5kqH;sy^xmS1Z4 zflGhz1z!qh{T`GwS^aZKlCc;;sqBW%K@>7mC2efmIj=geC8g`Vut zqsG7uP-H}eQkHiQ7OB>JeUv)r|bj;r`%tT4&Gq^@h~Uf4>Y zbgTgak9EBfm~`9;s=ds2dAIfzEmiTV{ouC(KCgTiAla7OGUtoOjXXV5pT;_o<8!QI zu+pzyZG<@n4N@68@jRr%HO=<5$boGy&Wh1IwgDof>p5zM9kGM;1M?Z%^02Lpfb8Cu zLDkrW?+rxT@s~6i6tEmEY}8W^t1Nqi;0%w}f zs8x<5R7QGWdnWx(k3c)m9@Wuv2;{!Ewt%`Azr|s7!r6NKUgIKr^bc)+!A7|#$EpWs z-_O(S@C1LmIIpwcxe$pdb$URL^R{Q_IVgCvr+YDnPmO7NGmC6p3ZdTj_^;%-##oDY zSp*@XwmzNr!q^sf5ne|fFG&_AV4YxDdp@YOciB0{^!HP6MB6O#lf|C zZ!km(;md3f`|Sd{guYZfhASMrdN-ks(I%BM4Y)Pm=MUZHH98^g(At8nR40C%+_Upd zN33HA;^3=QZ%Azq8I69e&*ds8^2ivZEuTx@ZKjR26E+}?0SmuD^I$Tdd_<1sfyLp) zMI?qfQFshB4AD%sTI0FqH zARuO5Vq_J$*V}Mc#A0dh|RzNnDewzHY zIo3=Ruym7@fwRlUC41nv`(m1(*Fgj<-s|)su@n3F4yiyJD;Yul7X&OHY+q9=N@8BS zyjc+bsCN43L{R8eMq`w%4g0AK)jw1Cj^(1u{l{-NEVKkG?S1Gur#|Pl37aKeQas?i zuuOLRw)BagHO&SaIg~RSG?%HK-1tzm;cV_mR-9POj+<^>T?GYQqe+VRr#lrd zz?q|i&r;|3uIW_67Y?u?UdLu+Fi^z!vwNXQ2-C}Wn|2|et|A=eM9zc@T%F(Ecc_!; z+_0uVD@xe8XMi?0x>#^^;?Y{#i%!(6h#oL1kC(vw6{WKN_1s^J9B4=Dpl(iaYCR$lOAd zpAq}4XFL1D7P6u1P{u<}Nq@wkp9p+M&|74basZ|ff2lLb((v{Pj|lwP>rKSMr5*gN zW9Y#FXVrV9GKJq7XQkovacg$KbBVjy7*ZlXtWQdFONSR`VzhN0bJl7XV}tBHXBAwt zEkY~Zrs(k=8;a*mG=&*!;~~%}=F20Pcm2bX3DnQ`DPp$bU1C8pwPQ=e#A}^(Xy*8H zQrl&Fo1wZpLABhT2 zsCjSh$5s~1>rensSuPEVc(NP&$*nc#%R1+Vi~&bHc*ea&XcT#> zrF3=tISNf4dp22T@5=JTlk%^>*+@gGHx?N!DT_k&EXiP2uM(7_aku;E62eN0MnLX5{{mRT}C}?R=JK^Dy z^jjSrVY(anSlj+U8$eItyI4LP|hc}wIwX> zkf!~@5&SS#yp{aChec*AGrjWczYHv?Fcb5`k-E#Yz`aa|{e$(-GFbZ@ro4)tgs^HR zmk~0f-kY_-^cc#pxi#Ko`o!D`P29K!J;=q}nI_;Zl%HOfSW{wSF&8pA6fG&tuPCil zAq4WJp7EWeYTI4J?zFLtDA0kp`dHG16}1jDz8vvNI(UKh>JV~-(zh<8{SQe{Si&F3 z&QsSk;0=Fha;#ASo`m8lXOHwwbDNM5;lo4`RrxRRNFpJ=TXAPK%HA!L;1!I^#x5#M zzp#lrIx*0mvAkr|v^|*`o2s)RdizPhQ5e*agGP?O`fbOzTY@e|;FTf5aE%59@x>Z% z(o1RkE#i?@@1u3i%Ic%ZcX8`+1X6Mku-y-xhiLTDgn zgcu0}fqJK@P>^7dz5H%QR5y6)^uC=xQtkESgX7fE%eC{f2EgS%Y`EHjF zt$ZA0k*)~Sz?iq^z93E7l^g-*(BvzgyU}d~Z&TWK?PH)i-giBoD?4)bbmQ`u50|}> z)=tH`_aC!fg&F1DOGpkXejEOl@oFvD%gx})%ltF7?@zPCcxX_4BTRPSa8+lkFC)oZ z7R8uxJP)%~!1K?-%~JK9Tf4`JNJsbDzTOqpjK8weroblgG`DV2PgJ!{Q1=Uk81 zkEaeqfTXl=WMh7m}~w>42(?^V%16{O+pQ9@*-JOV{QtSRwn%g2l^k&WtrF=)5@3<;yGKR~@fg zaN;oEn%tRQ>((pW2HqO6r>fG+Ou}5Y@`#zK_gWSUGj4`DHA;C-R)*)#O_$B()o0lC zOiP*x(&+c(JzVu1sx$^#1?D8vnmng!&iDwQVOuhM!>8M}*|Z00=JQ%JPekP?D=Qlf zwiWKOZF?vA<@uYItfm>ONj-j-w%azf#zAlMuCFp`FACHc-GdhlOPMw0c!cIBR~ya_ z2E9nc)p})5o*CL@QS*G*AynK_$797}v$Nx0KXpwx7Twt~BDVp{`u_Z_k6Topl!kfD zPDpm#GdlC>^_9T1u8OD1vZFOd<(UpWW>dPE_NK$#RYGCzgM~`$FE1)p2mHOJBM!~P zi`jL&m+TC&Yq@U{7JXtWQYAQFjHS0~!J?nNU%!4W$d;cOt|T{^o^MoGm1vl|ZqK=+ zd(A#RQ!UVpjEmcu@5{@L-&TDeu8>eip_RBuzjS01evWyao0TZtR+nr`9F^`_J5 zW^8V5?t&4Pp#XXFKmQ!eef2(6Mj~f!rax+|^Zor*y>+RvX-zl2Fq%AfIdp&Y_ICdI zYH{mdt5lzUCzfNQGelpqTp z2>mR_fl`_Nys@0IY^5n7VPW6PE951YeSWalbmis6EByTUWb$!y+!Bq>Mz~dl*_nQ? z!&9TZt-Xw<1o1XuAt4{G&4=;=H4W~MjmeGG?{OXe%*7|;coXYW&>8N&LLE-kMdI_?>)JBEzFu0HhD$r2e72BoI}>V?zuq)Bd#o|5X!ca1LFjN_gN&sB zSK-6`m+58lQ*jkWO3KOsSYe4Bf`Ur^T^}ADIH{wf)3xiE+=zNznp%|VnzS84LdvJ& zbpr3)xkI(iopG$l^_pYaG}smk2aX!@E?v9x2zj(Gn)saRcioq?IrFqkjYgw2ny=<6 z^cZWD<-=|6^>j+S=K!4)Fq0{a$lZW@QJ&m#yDT^{|Xm4b^FSbBQOT`Ol>lW4&#K0X=S# zSg89On;R5Ga;FD}1riYJ&t5bj69woyHHcZtm{-RwVSCI@beaaQ71p5^-xZF1uCJ}F zZCi8t+EdFpY)f_5K4sDDy_T9Sx+V4@imR zjJFJ@^EuW$$}iA}eDPv0S+%;({fqh3yN1#eltnVAatRL~{yNbXa7^$^E-hfIX%^o2 z#UT$vBrcbY)(1UAb<=k4*uMP;n>ku%F4pZ9UlpslO;|VWsK3`xiG;a>Y_J3;sk^&r zowZzbgicLBj1JFw&5e0!)}`US7p>qAY4pOX30XKz^)jN^L3*~d)!f{gH3^7tmri~u z+p|}C5#Vcr3|viJ=c1#+Vl-yn;G#Fq&W`0qjfrM{e!N{+IDTM2jAS|cDP(&;Wb8Oi zFP5I2O6DIr!Qz%Jh1u@R;roC&2_srxsLVmk$fR;S*Kp>i8^qU!7Z$#emaQG7S8a=g!EGK_tjYD}4-ZpJj4bke6j~Q&}_mt?LEeaIs zsb?WQx3>6*s`fY8mC1=MciLS2LuW?fD^jS!y=Gkug8da{B^zc7k9nT_^rvayh#Xeu zeEH*p_Frm}UbBQY&57|M6UWaV3AW1I(U8!+crn?c;@0yURUSTjw%v20*f5{GnRWKa zsnlLOlh^zgUe0SD71ht&i}lI&*4op`$h=a?v-kWj*Z3T@T)mKaT&5JJzN9!$c12ZA zY3e&@q#DJ&e|LLxYmJe=FWH-;4NgtcV`{gN%i)B(xYT5${uKSSDZ5-09u3BZ*#b6Fo;_m$1KN zbcU59C3$P5G&kZw8b;&{Zt4wXMU5uuf4p`9VWX!(YFeWC~$@?oTe;F2vZQcA=JhQjA$w;sLs z3yt2!r>UdEb4+$XF~qUXGzqUb*qZNa6l|Gx-N;`~+$v|H!#>ScV{)`NILxg&PFo{Q zAMagYC66UL&3E2{MR~D_IXZpgWGmd-c;GRWw;jpUz_S zdyZw<8(LZi>i6Qh*NQT~{{AybmZ~gI>znKMY+WmApcl~zSp2ZD%I^A))dR>*k*Rwa zb!M?=-`-f$UdNp3%Q6kNYRIg5)}%1~X+_sArv?Yj#vHQstWPt1;}gW3Q5n03(}zlB zth4hk;OZhh^qoIl>8a8*zALPIS5tWmxUr_W^PPTPyk??-m~@A=*Oad7(Ol0N@pqC5 zNTbYx%)WTsc;Phx&wC**9rwGMkVkY9!t37-%TIoec!F?hpOkL9bzsoX1s^cj{gb9+ z)^LS#tlo6uv2b})Hlxuswx}S>zN>=CQ`0k9mFO8bCYhMtazn7dvF-NZQ{(O*HLR93 zdQGr^^FR1 z6%2Fh%h)a`sMBAdug0x+VSoLFH_N!VY??1E&Zlh)UH$g$+X0gLfEQMCarw(no+rq4 zpw2we^48eFgn(o6QXQqCyj)yzuEU%r1}nRI)z+il1FZ#z74rZv*I6?M4jl~^7w(ra z0jRe@62K3TyIQU(OrH+yoO~lNd&_F9hn1+>Is**8jx3|MnCA8n!1!{MJd!ZZG%{TV zJ{(x_)<-mF9hdnzhr5A+ZwXG34viHlm2AD4j)g&PM%3B>V<4u4T5xBy0x@(PHBL<) zt(iRuKAmB1B06an>j_Bwf*9RpQW~;O&edKzcKTU_-;FYbIa%QdweZfPbgG|p_cP~~ z>)R@F`MjnFY~F3WMk-+olFE(itF}e15ZDZrff1zPyYY%JfeXL8;Cf*5GeN1eI*GoSQ}`l)HuZ9xe;_u6D*T7e&5r8voB5KynKV!>8v{Uuz)As>zQJKQ|<(*)NA}k&`hJ7 zYo0gPn$j;`C`kc8RUgN*)}RE)N59hbuhzd~9#5sUvdg5yh4JmYTuCFr!iHuz2z0!d)uh~7>3h1Mi zdafk+BYKlwKd6!EeWAvj9xU#f5lUq4&cWY?J#B1kqz}lRDLL9&Gz~C8V|dN<7_v-5 zQqGkqDNJ$e+n;zEUUFc$6m`)up4Dg!t|JKxlio%;G%*1osBUGkjAt(@(4n@Oc;z6G zUFb~D*Zekj=6*{{%Q%wCy~eWmO9tVc%+HT~Dtbw7iaixCxS?2_ZL!L!(ZwV_->|CF z<-mG`=uI9mL7<$(V_~uxc(SD7o8;#ex6smatTk>Ai}e-@FG1%Y*|YIde?Gc+gYEV)P_C-0Gc)*dZqd6rYZWb?%Zf_iydosuFR7?>GgvsfkSWhB4u^?r83 z*%dv=;Wt;8MVx+f^@Ntzx+_<%l%ru>VqGZhb2{Fiq!|Kzvift}GIp;@O>%toIMp^h z*!K~VLpi7lt@I10kA_I@BK`2EAz3^ULUf!#8|YISD#LOqhAeW2GiI%oJ#iUG$y>`tSQfde-st+Thv_8DNJjULOv@s!erL_Jzl(&Y5oeHFc9k{kUu&?R# zHNKWFDP^ktiP_Vh@vkl~j$&s#m8?!3&DOwO1EDyF=0BM7*b6E{Fjv8fIgo!3RM1U6 z8F8=`{Z(D=4rD2nxYO!gd*X*Y-(>)&2+eS9VK7x4mcFvQVR!fL-F&o#s~7Xx9|v9# zA*DmG;01DiBwI{7JwB#v=<3(JRmbuG2REH5ar=s`=!YEKnYv@gj_bI=-z-zo+*OTr zQq9)jl6}3s(eUIwp#_M)eg~?rI{YGoAGq9rFj@z}U>Uu;wQ&G>J%Ll#JlM$-9c>_9wSv%qq@hBAGhkzKWN}JGqTrP3RtE1ia_cJE;y21wJ?s0 zt0B`NR^w#~@`Ofa|3qhanT*rHJx;7Z-BqJ_@OHihNjbJ}*hL|>MLW7i)sD%f?k`ig z(EQSj-B#V`iHtEg>hPszmm{hzDjFUYs`a*Zzi>G>N9I_g8kos0d4GS;wuFq1 z2Tax&ke5F?BMr78qNK97==V>-);UrQ@<@r*M#JVz1a6mOXK84xWz+N1Z%9YtJC$g* zKdrtj&uhBC>*KRrY_CW*$f<#?OGnKI@7W3T{Poi+tT%co0$kfId*}bPrWTkkM$XG> zU7BNgLMmx-kzj3H6&UXebg-FK7>k$jsP|3?R-&g#Crs5;ncSEjCx!_BDL?gP6(Jv{ zdkp6UC6;=r#aNv#t{2#4t5Dn%Q|P~P#HG|B5FqQ%g-g$Xg(WR%wX020-OlG)yy2dc zrj1T7LG7J1{dE=_;0YT^*B6^Oy-v5!pO!Ia2?3xcM-NVd+-wR#!k7Hwr5 z)S4*dl-7gH8^EweIXcDw3e&bU1Ou61Uu~%(QWYOdpb-ehfW|p?)9hVM(HD_vpq^{2 zvU)3~%uV>KqfiGFvmAR{>g~#Wk15RT1+Wh7&4_uf>fj(2%Zd^Fsw*^h*dN?#5ZentF_Num=;2_{^%t6| zrw+#u9=vW}k`AYK#ukAv)lNvT$a_%x27BCqw8T6TRvV&19d($fVLc8BneRN}V^ZWV znR$ezpp2lRttM?qgj0WGQcptGq0^^NlXl+b!~Vr@(UAsgvNE@t?bc08%bA_*8TX$X z_a9#T{?UPJr=*PJ#$)2GiFpoihC_GfJtIFFgB6v~f;HqOWX58C=ZCMASv`-z zvvH`@?2ybSPpBlvnE4m$*9c1d=&z-wdb&5R0PyVm<-hi21UXy3(wJ!x)s>M-`#P&b`*$}&9McZN@s1Kjg(u3tq^%m&+c8*R!Da=#rP zXf2-k<3+2VOk}F5xtT{3caI>bZ!n}!Qq$c=zpQ8JJLv}u4*w9G~pU^{r6h2Jx0-tQ^Pdwpf8xm=^lQ&3TferNRw z;0>{_pmmMbBpLN=_){M+*9My|kDw=J#g9k;s>5Gu_ek4+J|6Y!jW=gl0h+xwnrG`m`Xbd)Baje}-i&t=}2vO;0{LaIV!n?Z(6 zfA-j^sf?JvxL;p1b3 z1k%i^(V|mbmP*9C{WJmzBZTo=8QE14yQ;4d5~FHW>9W`65pH9Brt32Qc=abKA@Nh= zW4UvoPS?jpJqLX3v+$4z6K0(=aZdK@%LNWPRKc-(1Us^G@FV>X<^wHncUjiG@YYfb zGiu4ZjwX2g)3qR>@)MdFPvy*Zzez_Y7kMl^XMbrnaMTlaPxZ5eqgLtlXjg2P{@ii( zfcb)iB6aorhr7*9)}1m2Chi)2bd6u9HOIUq!e2fntpsF4fR*N{oYut(3~CPI7wJVcwm3kv`AMp1XRR${wEElN%6 z3a{}4u7qasd8! zrfCPOW0Y-uj6oPG+oi&mBGp&#n)f1v9J-^hmI;~NF~goHqG!;Di^N2IT+hWtVrG(2zc8*^6UY)tdF+_+xzdpP zn$@R0+O9s7Xs}oF6$#Bj_~B>TJ?8CK1I!gf1EXp2-Ko`!PH1GHcfn6w8N9DI`^%oR z46bJ-CTpiJ;}ySYIMc1km9%xQ2ddd%QE*%XL%t8ln$UmtAT5jN+W;C%g$|rNd6Ld8 zX%)vB4Bp9*Eh7XXYQ4tYtIpv6CEh+S$^c0K(#;Ok&;({$u( zM^Mwpeio3I&GtgBugVnf#pXOy^ZazGbfh@;VAxo?jN_MeL<{VxZz7T~1Y@qGb9dX3 z6E#~Yw&*|ykeZ&10UypXXK;ZoLB1>ShUB5}T$zFt_Wl`1+j9}6ni~4o z8yN2hKyUwao*iL<5BsNz{tEs5n-_pTFij3`TiFX(A;}Mt%Uf%xaq`s)&*zOsp`iPw zyN-F?Alk_6aFl{Zt+-RZ71~IHLSNqz9nXOFMcrs_JOV8tO&5ExtZz)dAn>Qr#Od{4 zKed|`X+8z#Z1Zy7VnxL~`W9s~^>b%FK085J>(5{dGmmdUAvm|~big>^(E~+AS2r+h zf)-<6->le%H8srw&}LM~gAcfu)&r)11PsHAP$=l#D#XF5nGmW&X-XH8nk{~}!#Bli zZnhBZxnaA}6zE|3?aha`qSERkAsXYxu(u$vqzD|4s zy+lNztR=w{s}Kbtx^)*s#%zRT!PV5cXT^1;b0zz~$)jwO1{#MX_mu+Vg5xvh!%;L?gxrwE_SI*0nJf|?Ov+k37oEL0*5L83Wv z&)f|Cl}o2JD?xuZ2&W|ic0(N{m5xN|P3wo|w3X}j3$rT09U>yCn>J;%(UryDr7KK- zeZ#a-7<{*RO`0xLt_p^A(~Aax?7>m%cdXsCLBld`1LsFY1q>kdtG(MPzaS~`)VQgliqE|!a^AR zkj6`=tc_F}W7oK#Ru1NSRB32#A>;xp0resSKx(K?HB>rPwett$g_=R1k8Lu4l3*(Vr04MAmYKM8pwp zh=C#rwTM_Yj1a!2GbP|V2yGdo_g5s8%X~jRUXoEVf1ku#5_Xsq;Xr#xcN-swnXkoz zxB=iQhSGRTQ`9*gbanw@(DkJ}2zvpVcFCAV>UcSBjvEwUVsih3^T4)D;Lh??bCwp8 z+xrJ%Ld(H%oJU8=;cUT<5b1|BwS+tFEUEL=QY(F&yfA@B++rOh=ri{HjdJlPrd}-; zP?vOjv74w;84n7pq##N};IfI@t|Z=8np1z=b2via?aNCGT0u7Wl3F)!@rGXrV4D77 zqJWFn$%Ed~a+TMLm@m#ZGA9;+S-W*$`E>$j;@5lRuQ$mv4$@ysd@&7%7)b$w5S3eDvkH9`MHn6nbPUaKgKzgp1(^8f+RAnh%wil7yuv0sk zN*GtOf?ALR_t5WJ z-pF5I=JVks)JJDvjG>||*fX}N%8h(9o8pyscP9Up=-6u-6`Hm`zYg76ItY!H+_@@r1f2AlUS*!`?Me;O$AB-YPSYx2iQLciNa-G+_-$ zfXz)N&M9$Qy91E5K=V=#kAif45ENA9%5~_IydOV)?7KS++u|_xUZwVHlsrSimJ<;x zA7V(XaO&6Weh<-j-CXo0`Pczro&pAt6$Sw2MW98lBOVPZjW{lLA*C=UYd0VEdmlTq z49HJAy8o{U_7iLk!nIx5qi<)e@?b9w4ie1>d1WOfI&isqEfKPVz#z(`^7J76Yki6< z?zVzjVV2ri6#7b+t&4I&CwW5pOjmS54ve(WYx2NtVtUxR7vm-y@uY4 z3zi>&?yJ^Wum|iqa~62*RA)|%8hBtm&xv*o`7jjfZQ<^Nw8kcsSc8Pby3%oFx>!nU z@LtoCy>_mpQl09O#lNvois_K-G-{8{&CUd%zHK=y;9>tXMc&TDka%T|sY-V}5u9Qo z#9~AA4Rtg2I4eM%YP+MR8a+5ou&ufKEJ&p`J~p<-a9RkrM(C_1R<(VSeSD8|6NoVBnXxmy@zzZ_ z?WJPIMR$mzSH-W3`amMK%8&0G{hala*!d5{#<0QAbHbRkPc|TM^0B4(P|T9vESJe6 z93vE^KXz(8f0z$Q2MT%MquQ~z;AmG~UbO56B#wim3(X#>wt2s+fFtY>y^L_`sXXR{ z_yaB{9)Ub)WnrcFEWxT1o&@(00YFdoM|#Ht^Om>VhBpbR8IH286CS%87f_D&MG}zQ zsQKk3j>tnW&)4QxRi!^C#K))A30h%;5uWp=fPy@D&R6G9m|@?K(uY4-&&YJV;`B@RRDeSA*M+Kt&^DE7=ZYpM%!pCG80O z8#~|-)I_M9eRqpFMHrRPsTftPz!_*)1dVgD3Hsna;1AnQ^kxzR9LEt-bOE@BV5j=4 z4|<71>}iSvvP>K21CiApNLVAdhrM~N(l_Ffrlj{}_Qc3_p3=2kLyS`-yOR_^%zdlZ zuFV(8`m&N-K2W6$SpY%qa2-v+w^oz@UtIrNT+9Yxofl-kauhCTtT!MRzP0uo3KleB zCHfOv7R(HMB;yYEvL%T01&)p6ZAiF6L{qbAbm^=bpT-s}4DQN$)TB#jbogiXc9ew+ z5Tv~U9=>v9pzYhYd!t!Wge#%-v5&|N1aLmFFMN7@n~S$Ibvm*K1;!nYpIwe$YKVGI z_iSRIKKnsvUI)&~9exWNG@y!LI3GilH*E>Oh~yvOrhOD2dghio1(cUMR0&Tb;x2Zz(;X|wqvPFO%5tf1CchY zK5%w3QSmB!=f>t{4dwh$|0&+TUFLokfg%mnyR3clXBd2cK z^cZ<;e&bD;kvPsDb0rJH^$w61Z`uS3iwEdR*JC-ct^T;j`O?URo_Iz+d`mpW#>O^a zqB+kU5#>r;4FuOaxB5!7%ue~*}W4?vhz%ZZbkB%~m@ZQrq@aB2#aDHor# z{j;bO#2%Y*=RBoa8C+*88sHm0t=g)H%t>6VWb+X+fw%}urF-I>0Uul71m*5K_W^w* zFWHz_CCN{Jt!jB1WN;&VZn|?sZYSQdd5-lfjs7HOF?oN=zTW$<{tNzF&U^e57Tzji zw^U&O98U7jf61sI_y7ty7Z+*awK8mt<@|r7{4w;AEXQ`OJ#*%Wk`Y;+i`I$pjK)mW z0+@!!4+&x`Vb6MdYbBjID2FBgV~9u~%Xn`rrX(v}Cy(Htv=RTaxACtrGyj*bLM*O- z-pk2#()Q;)R?MTkONYm)YU~%3u+sO<{|fi@fBAy{U-5wdvsXc4@cFKIy^P(ZvP17b zS4JJ%;5q&Y&QYjXXw5Wg%ZTY>8LwwVxF89oUoYn*MH@qS(u16zgwQ=ge^TK{YIht; z-zvYK$&3DYGk$0hlaK?*W3n7Rmt6{6ziCqfO2y}zq}F_X9`eGMi-vrcI0sO#gAgDl z4npyqfJ|K+B$_yQ4=ynD(*Od@9an!OFMs~ZJz}l48!RU+-LZ)Pg&f8n`IJ#K3i%n> zo!IHh0i;#-Dv@8fxU7{|>@owsJ2aO9hd0q@;8P z%}P0Zw)Z5g8}m2pCYP}5?Ke==rLcszd7)D3l&+*em1uH&Q;v)_e6&s{Zgy07(1UP*PL( zMFv>Q@@%&;TQ5;0hVYp$CcEG%A&{OZMQhT0ytR_g(g1yOD(phgPGhy1ukwGUyu8{z zS4Eli?^nDynPgTKBP}rHD_H*g5eP**wpmCun$wJBY$C?G zdQ^nKK5O*4@NAsF?7s|4|7I?BUXvqHJai&oIc?fEqXqUN)PLsGVkV8Q2ueZ6%}L)p zyatuyRNY}zUVD&76*F)pdjoNg@NTD2czx$I7v%KPsWgViw)tnv7@3L(!sR_$U~u;a zT^RFvhAKM7rhTf^|iZAxfjCgZmt+<7^EJRr2GXwuSt{ z#RXJ#m?&*CQ&3Y|JA$qK;7Hm|>LH;yJeF~B-tS~nHxCn(hdyi`dk^{YtRXOp)H9Hl z0))&0N>wDjb|`g;{iit-AL^fV`9R-qfS!rk=9WiVYiHPBi0O)W97*T5!T!Vk6UV>C zhqZC`j75G<`MiFjqoFaI)3wYZhF^FiGk-cnNa<8cMjbaxspWb$gm zq1NcNa~`FJIqoHPajsQ{>!Igf5M`wWOPnQ=(Z~UIQ79Uu7X(CjXh6nAc*ONqtr%DR z;T?Y}tT_1dPvE(K$I1VcIQiGk+5MkJ>;K=cLjBDo|I7RrU@AFih4@pzFKX6V0Wooy z6*7qj6h$f;yfmAMHJP~0R3y)kopbTQRQ@6+X(7pIADkbXXmlkuW|8vnnlqrFh)?=3 z+Bql(>v!S*Bw_U=>eTNBl#6rz*={QmvJ!?AW|aaWHC-d_oK)(%u>MuCE!TFfN9uyy z8F!@rh~1>5@kFRO**p1RR%yDVRcvX6{)u&5k8G3nE!Ms#;_&kK%NHMhSZKd*yZSF_ zw<;<=m_G<<68KqrTm65ax|6RSGv*a zWQL0jRk~mlsS;?Ap+5(+)H#e85<_d_N*i8YUONEUtu!K~F|rq|BftGn$4v0+Y@btZ zafp=N$!K+3P#~Sa6<+KSG8Ra00|DODQUjkaXZ9&0Lkhre``K@^qJ7VzU$XI0@8N9% zZ)FLBsoP#SedMn&Uyb;@j*FJ99Iit4^&Y0wJRvx$*8P60fbz&3oK-uOdtJ zGOIkczu9A3u!OZLozYOtfFL&U?y!7C`@q$!S1o`MEz8>-FwYVNs}G~NwWOp3a-z!E;zg8) zzcj|q%O>KF^z718c%Sx4bH@22^dzH#trNMorsmu*a2P0l}zusM_flKCLYfx2U-V1x^$$=mU_@XAC&h3FWJwt3xxN zf~UJ9#O}RfV#e4seE!g{q(CpK(jk?^8l||9)wme;gys3Jj(4`*aCQrELKCH<9bT&U zFsuv_y&t+_F_e8W+-U|{0e?P38wfZb_SD+06Ybu%!GYC|hO*B&E<0O}@Lag~`$R>9wfP<7lN1>6HXY`q{by2F7B-Uk zQSu`bg{fLX`t!jvRiID`QmZl9advt_YUHDuTD1BN9?pQ-!;aM}opp6|cAHi_d6Aa3 zl|~|rn%I=>2IXxVZMLs}Rs_w1{&MTL16@ILQw`TG8$& z(|3gK-lN;3q_3|}Zv*pM2GV&5HWx9PErh<+-@R2I9KhPo|JYAIi_*vE)~^bSwBb}8 zOS+o4TT00+EETa(w)@=sdw;L{vVP`SHW^ic zBS`97Qg$6@VW1N=!<-SRo9bH4Ds*t|au|d53@X5&cV&|PH#pj_x>0wpg|H206aB%~ z5c_sM!1LbB)Po0%jEr=2bs4OkD6~(L^%FVacRaTK8>xBxWJBg<5=*Am*8RgZD9E;z zDiT|0Wcbs)*W9oegl;%Vi`d96u;1E2#WpUaV~mQZpJ0aiOTRtyY$g4t7v;FB6ZC_T z|IUFTfT>%x6zY_S!tq|s+(`{A!|h887IzRM@Yhdo7-NzeRrU3Hy;A3@rxhNOu z!Qfg_4^y7Sz^eZA(WjBA-09^@kIO;atB)L7Qv3K*l){WD8GH^3e%b>>iHx6v@Q;s5;$F#5NRMx1Id7>wNBI#Lc%n2*uP{Nur! zoQHEfV98HRl0N8d@QQiJk7bl$4&!O11E$8_LlV2EFw=F6Mi+JewDj-a;JK0hnaIY( ztNY#`Sz8Z|{|sn3wkQ{!FwT@>4j~y34C%ieGe}*Muu;u4?LjgB{_3a`u(I4D4r3xR z(a(Y@)_!rmd~Hv6Pmf)R7^`hh_{7l@5hT1QAD{f5hSHUpnW<-Aeoz~<@ocNY>}G(3 zGcdy5p{|#_@JFO-=mRo*mdG5ukTBx!9WiA~p4Ce7ta=Qac}Ow1?F%caM{AA5PERxhC z22r+N>xSJg0CDOv@>^}7#;3F;`FP@V_%$mlcW;_B7T{*eDR*uCGC0VN|DbHPj}Qq$CAwP)ZicbuVe-tU3hk!CW| zg#9Zs%LI6Y#(p-o)T4Dl9r-!M+3ku&Ip=Tt2FMyAuDpvmpZ9euVjooP8XIlS&l}t1 znjXL6&*JOsJw~u*rtKnh0*;Z?tv_nKL!G^f8?kDoQsI18c-;iR5kTZs5 zWdenOZCbBFC)+s}sepi9l-dza2*0e}LH#PC_54@bZkWV>UiVwGB}-gf9Bx1<8Xc1m zso$3!;)$fQ7$!A51l`l0{O={1aWdMX(xtbzm+iti_T^%o#*yX*ueq53D!AO{Q}JG( zA6=tUiL%S}AGi4gw@C)1s1x}TIk$0|N=|Pb>~`&FD4;-M-r?_@K(Ywj+=!L{?B-o$ zPHG6gK&MPA*oG$Zw#naysW7#pxE|9SrL=O#QWgw2C5Cjcc zglcy$P)IW}I(i3uF5QwQU?SH;N&W8Xtm*EgVglahjK3A#N+aW7#0L_ZfJCc|(*rtJ zi*Z^8RM)dUS0zZF0f;l6IB<>&Lzan$l#8=6X%z;f!7S0K& zp1733M=LzGBOUMjyVM?O*3b3yAcTLMewH}RHjnGDMN$W}$#;<5=uH00+f+mDrKP2P zpKYOF8kal}BZ)E$Js!qO+O^gs=c(@y)f7BGR9fWG`^g$Tcxer_5JNn=f4|`Q3T;=^`29` zx^ZmGeR0q_1c2>-KSF|*V8|Ci&Q(%VQ+x2>fh{DNNOZ?QGR64CAawE4WV{E~ z9UXk|<8eN@$>ojM_~%faGvIB6v(TR$r$WG105uo0ibDXmq9osx8>yFIY$a9>Dc(9F zHI?&L8c$Dljo0A=Csy+~<8Tc)fuyzw3k!$x=b{Bj!Bhzn^JW+uQ&^l6s9t7$h`nsYr_cm-;9Y5Sg0Y~`6cYD`aP%MhuTi2{~HUW;sWTf%LV7nBy^$^y_G1H2WI2ght%@>-Yj+F2ji|^8CKiDmf zMV3Z_5>t%N_vhtiar8<;M3QD%auQm{43RC6xiYiZ^hXG~vDkaNmtbhI@cSZy^Rgo(Ri z#P~$)=UZtQF%f9P`2oF>;_xRC2OQ0yDclL$x?%md;L+?l)trs2@5+0m zG3#-jQPsdmP|omEVZVM{h!lp;fr@6D<{Co3_4}@YRtp=^q;I1Q-Y;|CI)jX5UqVis zpi1GKuH7)&sNKc8>Lipk` z_k3%Q6D#6hW}&H|77|8FZk9B?05dc?oL~+vp-NBb)ab;-|Lpe|^ZPNiK&L`;HH#`} zN`E6klZUg_(IphcWH&7t8`a6=TLm(Bs(We7cg0~{45CIABw~n>T6kD~@b++jQ*Ln& z>Beb8|w zfK4@!I)Tg<2JV^DA336G>Djzp^R3|p!Bf*xkDY1^h9m=sC4fqLJMyBxFMhB6*nDzI z6FEW!n(<~b+(phx@iQU?k^@#sYr&MIpt1B*Kv8PUz=V31%r_W?ry%1T@C1HKuz;rt z+;Yd@X(>=bAS`V1>-R@C7Cha8nOn;9r zMtlO9qpCSLSY<1XIOV>eWnvp~)Z>c=CRsl;I`yDEIwKagmi+F+NP9OQ4#J|u9QfC9 zZ~q_Ad;k6j@;op8&7{UWs#~D0i%}JFSPMhn4raZ`M~_%uv7DFjN3t}F0n0_OWvciW zknc6?_l#T`14D6yu7`7TCO%w~S8q2ZKcS=G zCyvGIKNRjEQ;!K4${^aW;$9fHMpygmxk)Luw=g+?$fKrmzyZqQPmp=n+IRTcJ@yVMC7x((ZDpXf6RP*xo0l;Oy9)Q##=fp+Bj)6)~Tg^F&1s)7B0;x*1-w~O+ zItUs^g$_|lzWH(Nac`_zVtoQL;959giF?Ni@X%H7J8J;E`AEDLY)7nbL@(8r|7_Hsb zG@dUJBL_JN@el>YLxm1I*jcc|C74lvGJbabj0aD{%$< z?eMvFz*!qVx{kaJrT+$7jdKzIN?GCGd50|Zf99Qkp@Q>2VX4iYjpBCq4Ulk#;6=Gp zL*|s_=IdWr{OuH9+D{yHkFsUUH&Xu2oqr(c`Tu<{P(P5m_?ZKGGD7U6B27J9iUEpv zBt42;*x1pxI^=}6aG|ejvD?AMsJp?9mf4)mLwPAP6c(-3FfqG?xllLD(GpxerP&gs zNF3e!eU>JZk>EDd ziRm*sHa2$DnS=%1a%sYZ4FpTj4Gv&Kc7sK9h80z68-*6Qe zMRMwG;X-b`TbzQN3_wAknKlMUmV;pSE8tW&SGqy=V$w(kW0@=Nk;Eh@!G6c8+YON6saF(ld8?InINifxk+$ckv|?Q-FJ42FJF!W6=6X> zv=0C%TEL;5RhHjFoHt11!^dG_;qyAd0IdrhLBnO!!>DfI{wk*R{LG4)ni`u{HES1@ z9~$KnSG?^m?T^bA|5{yDWsC_fZA=A2u&J!6Nq_s`I|~VnB)7?dZ2J7xR#vJ2;*8FP z%?;~uY{B7Xct{Sw?s2^RI)&mIi5qiZS`SVM?VA{`6Xv0x#ZWD6Ra z!LX8V2H;dYKsoSEbSj}Pq6!2s&Q1MRSw8F;>3->k8UHP`wIX`yZc7(86uAYx!wjv)#1I_@AGIwmZ#)=^=qh}&mDd+xEQv-x%biZ zg&&?Rhj-+U9(ZWv-lQ#RMCgAjtM~uOx+^p6{TcsQz;g0$1W^84i*(iT-g%UVKaxy> z-4ZLe2n`VfGYNi)LV-%4M`I+HQTk8m!+E0)lWR>Qjvg^=BgcY4#l&fRA&|zSi&o8} z3?IdTx$g)e&*A;aq`Kg-A6{mi{?_3C6Ausz%A1D^qgv=lU9JLnHpL@>adE+f9 zANI`PNWDVXFX&W(@t2D$qWs-nQHm7*0R+5+IRx18lm9o6XWt_P={;mLW1I}!Wg6Zy z0WQrLj;#!!_!v1Q#z=Rf1W z#C`F1$lf-D!{m^loHJqyTztS`LA8NcujzKjs;98Ygz{`lXtq}k$VrYc4?V9kl`XV# zv@LwBQFETU|I4NOK9Lz1asuiExTO{m@A(t*3-%3=d`pgryT995slW)Tw%pQS&&FH{ z_wY-*O$jFam;*5Z+r24zbe&k)*m{atc=%ZVvL=&$o|vBi-QXVJ{m;z%SJxiK6j=a6OLlt=nCo8BTW7!+XonPrZxOx|ZjMg4ZXhUpp%cFqIVBwuWQaS+!eVl= z0-0^0(Fu`(A&g?UOokv&O$)P#Wsk4`ghdD6KnPJg9GP)$dVGLA5^iVSq?yg#AHA|a zmeRO1?L6g~jrEsnqvR10Zj4NSgQb%JL{@>6L_Sa!#|Y?#a311P!|~s(Fw6Xr#p;B! z?;YXT$-iAOQKXqe$9%qZ&;(gH(8lNRjA?Wtf`ZaiBjz6>lTG0KvFa{`vMNo(P|N_G zL$WtLf8M#+OL~l)Y=G-`FIXH(0?-B4P+KfFo<(UPdg`nY6Wry*Br7)ziPn82*07-^ z=oa`$X&eBYKWEXtn=6#)@haP#H^IF|e)q^!Lrr5k+L?fM;Kx5eTeDfwEtt+}%! zqA+vnVW0=c@7>uW@x+(N5$D>;>*S9@eh@X@UkQ&X{+>k>Em(v(XMcB{sOy+)KOU*Hb4LyK4N ze*_2p7z)AK8YlO=4pl=2YP7gmO7p=-e{xd{R$}nCyCI8#O|}^)KGROx+rI#vAMNe^ zV@bF|Q~D_KF6IdyDBaoq2o|4W7EAMF$9xy4yEss*0?Nvo_3L+R+ZN?sLZ&a3R8*=! zQGZOgF=`!$k6{fzf5N+WM(=84mSF^Q$vLZ&=MEM?>S$|gd-CDKaWRXU50VOA zv+fvTKL{YjB1Dg`sd%uMX$5u5mMx}`WyG^bs9h-3JTi`PyLazi&BGIeBaIr?VVbMq zfs|b&q~OPxaA?hF+mdev51LtJq%ZVUqv_^49J~mnFbd55ns-BsDb2>1e_KVXfHP|t zBTYO7I2`nhyS)4O^Q%9vKQldnvobSXM`ENMy4$`s)q}W>!EbigrbIEbv&EohSg@uM zi`~`n!H`nbhc0JV;XJ_mVF5!`PPaimt=jw@LTLuRQE6A~;7OmPl*OuJdbzop% zfIQMnr~a_s2MEONa&r0*nD=@f*D6@PeEF$;`!1G-$;HCt*d~-;R9UG8MJHn*=21XE z0M75zpO~1ybR!M#QhNsUg$EfVE5g4i0s917JvvO@Lx`J+t?ll~ zejP6@Eumkj(O3b;0>UYhewy;g*#hm){WWmm=h6KaYx%|neG$G zHhr)`(w_i@wD;s(x^&5rbsFMd5LM27{5Qi^-F9Prja%k|*B#bSRxi@oOL?i>QdpuG zSKrXEWoA;k5sPS?q}2v7=|byGL5-LrbUcdSewStZvRVqW;}O+~hKYE)Sa?$)s|4V9 zRO8{ylma$ehP}(@?HTpDJ0kkhEISM-ULSbVjlgzH+!b7+q^f$CiuEY#ixhw=7<8wW zS0r(lw$$Vrl5Qb*V`wvZgnFFh@$k1>%?mFE{S|SxH=v4I ztz;l7DtgUG+w10P#Z+3LpvExpvnh&8U7BUEhyn8%bjO z>!)ofwfFwfvX2LPVXQQ3$V?T{&wPS$)KFQMijGtTns}mWADbasFwobO8{T;@ZbR>9 z3y;}Raelyczp4a%-IM2z=_x2EG^E=c=<4b+t9ucByIIch%RL-Q@mR*G?=w#2?Z%4a z^!J?SJN02OIYIE9SYTe}63UVVKVghwS-q_$)i2kC>EFMi4w*0+aDJp4{iQm ztzCIM)cYD%o$4IwbQ0y1rNaQP;zAWFYZ2*DEY$U@6AyK$i(>EU+HW@gql~?Ik>NB5p17HN48Ygy6PENRtw({kQ zlnWQWBf@PY#H<$Q-l-K_H*~aHEOhY?X{6ECNlQzg^!9E-xQR+gU`}3#)`&j40ekBU zOYXyvv9!3jch`xb7FfZb+-KCdwn`g^JeH-MB%HD~YNJ&9{tOK4>F$=5k&#)vWC>iF z4XA5@aXiSmhK2@t*!$Ec3A`sxUIX>=536p!_*Qcq>EmIt5A`*y=~UlyDb(ayLKq?{ zC@6^gOj88D`BZPM_y+_c_&IM}`qD24_Vs!h%je!%7*M8zkaMWKriOssu>IQi_Yf$6 z=1^4J-|ldjMCo)Ss!mIfV&QxD9hUlbz&nQ7`KW~ZY`Miyy9cxk42_I-_^!}C25L#( z`{&-LVOk#67)KIAIR^1MZw5;3GZ1Fpx@ld#Dv^b-)AFI_{^9Z^L`6=@_d)?F|15sA z6Q2G@X^r%N03p6D2pxrh#M?_0<2ImH!6j3ypiotD+2WErwV&o6_ZDHS6CKqidlo$W zul`?oC)ylykD`bKx{?t*9V3MQqh721{99ULW2tVoY%5b{jbw zii~yGY!{%yvhqY;OE$cdAsgyF{lirUv3rCVxRszotGaH*AMiT{Yyd4TgkNhElxZ|` zbTc#|f#F7t8x>`YLkL6w1g!S{Q11SGsOw;})(Ao(cU$A|a(c3LFA9Gi9)5|X7#$Pi z%z2YVPYsEWkI(o1a1ikUSfTH@k?TwVNbTx3myW&2Olr3JV)q|N0LHjJ&M=yK3@L!? zG{z-tDr_gs4B)L9k~$u6$9v&ir_tN%?wirui2A!^#Hs;Oa*TW!2(8eqdf{{aE_Q8y zJbk{f3gHFv%Z*nZ{4(&x8sLYHj*ee_$I>C<){vqv0%)#OMeFdJamR`>xn-JYZEG9f z;lp1}Tq|i&FiptDUWX!uqGy$9841I{ns(t$sCf_VkKW?YA+d1_&00%K>tL?qMQ^-& zcugfOEqf@QSfM0E1r?kwA2mU^wq+f_TF*c7bTv`R+(Epqu5Lncz-g>+#2LiCbE9E+RG}=0&%8uPZOC&mCF+5# z#`|B%L8JHpHpiNwD0j`B$9rNxhlrH;%9Y8;q)`sVTP8$!NG5H89{Z^UHVG>URHk1mZAcxkk88a zO&7<%A}4-V046aF$~lhmbAp_4pb6Qd0LRN-;sI9|h|eeZQRre6InQ{>e+&6;E)Rhu zo0tV9NSY2?>R45+(0ikq!lwQ!^9p*NhZ^ zoE_K@!zONMo-K@UA&UsaH0SQzjC&4Sw`?x-r=C_-?EyoGq3G-BiM`LqcLHRC=bltZ zC@HOq!Z$V-qje0}pt)O!kkHVEn|rPZyEE}k9do_;o2^8mfkL1{RCw94C<@S%G2Pjt z0|25=A`$b`I5iYb4VMfJU=z4iE2zozeEfI^U`BvpBIdffWvpORV`G)>Pi;jvky%WK z;vdznI3FM&2hssrHqVeAL|fmc0{2#HNM4zPBCWA1i1OdRC-N{tPAWZrhR`FK;HoH;ODWi z{Ih=GKq$+;TCVQn>q~_&B7AU!(@ZGoxe?EgV&0QC0}x(ZN@_E5u#$bSo&es>`S(@^ z|1Tzk*E$U{;r=}A=k2@c&IN?LHe3@snIpE2s&c(?*0Dmlp|*00E7+Ol63Akn|nkUk_Pr1+G_-;aNl; z)3)g3!3Unnc*ChUX5F>zdr}Rf>mjE>5&{eiaK0=9H6mPXIu^$usvgLT2LzGx_xC62 z5o95?^?vwgywmmH=1g8Eax1ReI8m#e8O_|i`w|rl0tIgU=npAD72ToJq1lb8Y?AIl z&ghNpD9m*6(xn0%nFIh>n_+`Zvp`wzyzo~9DXT~4zPOF@yaX||kVpv)U50q)ibb+J z-pbH01R*==&71w|zx^hF1yJ|hiKh&}lU8by?Ph4 zVKYKnBhu1R>L>MZzxRsXV>iDEa!m>Y%^|%T5?^(Kez}QP!p5MoL_c?qmx(4iAjZ}k zGBDQ9SAk$U`#H(aUv128WMeI(npow2B1I_hx6mn?Pm?@Q8yu{ zwEe^lz&Dn%s%leuE!o>w*mH-{n0>$k2`e*Tt|K|hT)-pm);$j%JU9w#s&OY+DUva(<>cGU)xC;-fM7EWqS0;nT}NRgj=8Q#L&Rj=grN{Q?pLvTXC7fCc$M1 z_+4l#qjDKPuVDFO&OPQr)ia-+pw(N-iVcW%q8yU?;rs7jg3=YhlKVtG+}#hE8X1LB zGGPAd9lx_lj@sXU6!2+NYthW)Cs!0+li31C2Ez*&Hi)lr?Onn1WH?oIt*UrD=Bh4r zPEF5n2TrTFbKae&Gdp-0ScdETpVtXy*KDzG01Y^4R2HLTp-@GGKcgI(C_gVfa$u&x z8N4rw&6XuH&60J)C#!vB@>({5(oq-?!pVlud+MgzYU_Y^1>tCOKx_*~#}p{eyFTQj zC(VK_Sr}OzmQC2=>|#I8haAhMrbU1<$W=s@KS_@&8rhF%9~~K)j8w>GhevH)1*Wbd z5%KYxh)m0jdRFjhOL0L@4@FH)EtSjdN4iP9Lr!oWDWx!)DNGu2yoPMBr5x1%q(cGj z^TdR8uxP3B2=`xA+4h{}jTCcnO;j~|aw#yTw zJ%t*V4XGom6F#Zs`7VHxfynYgUW=4`j-dp7kQ!>QS}R8&X~&@ejSONT3}RL*4Cgqf zE!wTLoZ~KvNvX1gw~e-w7kDcCxdC=`BF)cW13n<-&$&n`T!);L*GRWMQIGn|lj7R} zqxqwd&6Tu=Y}mpe(mu#IOw=PqJ8s*}a%_)xL#DXc@2}!=0h_}qI6KKC=;*Nz<3pT2 zk^;rL-Z#Rst`|o+KiIoEPU9JYbe!55ggVCo4C4$5508n9voUcXw<9kG{QDx%j1DW1 znX2>f<7CAmXSDXO6A-K&@?4YyZgdnUYzf!)o81p7wrC3IaqZ#7Og}yg(?_qtW2V| zzdZ&|vbiX}mEG3~%}=2%tLLFjNJE3A+~eP%=i&BYvE&K(N2Eik1aeP?;hUEP1_ti% zoAtTztCu0UYLQ5o^KcqIX6$@aeoCJG`d+Og5+$(5z+KtfvV7n)$Ry(cDwB&ny8=t3auE>=AFlr`)z}bh7PYDlA%uEYAvtD_%oEizeSOWu&u}Ql7#~~M{s4LLn$AvuzsZkwSk~{SG)erTW=W#d zB(n>bE*Y}h+S(8;CyjE!TfiMbU196RXGWb1U6E+bgnL!g(69x9h(bJoishZ*VNbYD zm%YzdpE}3H^g{oakd#CqO#_FhuB$UfCO{ZjJw3e(SkQU$r{!xZD)fOM8-Y$5$>4xA zFiChictu#ww}A2#aqBpSD%C%4f?t;jx&%?D>t*HTjcLZ~5c#k~4dJ^F7M)B5k_zil z#&_pm96pUKbg(EpDB mkjMM4bNxLN`Fn0L{!Pel=iR0ayV~cf76^S4y*s*x0{;bTtU8ha literal 14286 zcmeHu30P9u+c?!sS!!0M*rGP$lvAeWm|2oC?TcHhrYM$9ST5j_g5ok$F4JggnNlg* zGHFFfYJdvjgyWW}nW-UqjoeZ|P(;Fw?;7fizyJ1p^E}`4|9#)n^>`mR_q^vl@BW^1 zU#=c=IHbO8?J@`iqJHG?{u2<$*Ek4d(K(exV2@9`fgc32An};psRQ6w2*e5kK|!pn zAXeZ#1%jfPo124O(;d?n$_fRx6e|?XDka4#1-x_Zr!TaSXl7Z(@Q z=ZR2O5EKZ8f=t6rUn$@pDh0gLP$^s#jfSFucLfL!0#8pceI=ojz(ztRRijB@>rMh^ zM}rOQwx50iUL;ZysXD2;JE>ZcVg*UDvPwZgQc&Q18o~7RDYO(C*tjWNuqjArdr}gb zM5-oLcaypmG!%q}vZAGc4MhWoPLBmIE{zK|1x=x-PO`5~LRXW(mQ)RbfvBcIyQ{ku z)r#tBg@T)6#Z5tRX<$p?f{@do;H3bexe7%$3EfRf>aGSGsk>VN0)e+~g@VuLgSf%j zL3jn1rT_uDNs8|3&kv{3f)j(#3WcJ9p>PMKu{7YYOE3hY8$bQOAh~DgJOr`5c7D%f6%3B`W-$s{_zif}s3go{UXe;MMmT3nAfY>XYh_AGgOq zA>jvc3+{UVvkxD{fV`Wtm1{Nb!asSa6lA6^Gg z=zPB{V)Mu-Nt){r4V3Vx&V5%q#`Xkvh2nxuw?-yDNbaLw=pM|~ASj!2Uh@oQhekeRLl4tTz z#T0TLmr$KGb=Io%iCZ<5yBxaCGen7; z+4TN;?ROQ@=8MgvWXlX&y6JO$@AS71O2_hlLuB)#LPVrXG0>m4G0A_vTGo2BAF27` z+Z#lWPu{l&S<;YO#%Sm}eqMQfIR169{}3^ggaKFy70e!Cb((k$n?@f8sG_RqhTEZy7jRYbEeJ@258J_x7 za$#Ni^-{TROR`IvG_<&(-Mhck(m7hv7GcV$Uz}W_&~9f9`+0j5R~&i;PwRXkG~!mE zgdc{fqgUX=FFy>Y1?4hOn9`OeGAsW)=A;7&mRY6-(K`g^b!1Qi>rb>k9W&-IXxGtQ z_LkKq9RkFb(Og_`O&=^{!*T+X+9D9rr2J5v?1>{4n~vj!OJ1eN(mB~(0=A82PxBHY z2_efDcV2E;t}jK2z1ZCLo?*%Rl zmx_|<4Ej-4VJidsrtPu+IdT-GnK@wLrU!V`qout9+b7<#Y9p@W1)3HstfVTAKKU51 zys>C`bvZ)ljiAUwN=|6~KF*DH#=-$|?CEzE2H&Cv?)zCQT00g2qrIIk3Wzw_Kw8m2 zb@C>r%jgw;6Rgw88Cr&nHs|j155q}5Mz^p|+lEM+vRjQZ`f`LJ@>umhD$FD9wbu>O ztKgoe`#y}9q<2o7n%Ya1wZi&ee$Z@s;7EgJqf43&4Rxe3&Oqs#S)|dZ%Md3+4UE)Q z7Tip!S(3oBjv|K)Ra3X8s*d;2*1O-1?BHlvc?`du5GOuwJJ9!p3B~8s*pUyKx(?zMdu!T-W4_ywPiU^^p56{JP+&+Qa<* zO*~_k1Z5k^Cna$mzIuOVOq>-yNgZwLDi9<6BA-at3E9_uFP_)e0zRBce5(!Ez7Hf24WM3WL1nWCTaU%ZC(}`Z^|^pZG$ILq-%|rVj>A-MWSwf zIbM(!Pc(eX?5%6EXER#c)&}NCkBl}#^qfsGdAb&5hS|?689D%G2QXap{P&wRz&TsJ zPi^B3PBKgUubjwUk4$v7pCSi7&gvCah3xq1Eq7~rJbg0eMzj39vn6+}HrB!@+T$bI z#6=GdOwjpw5)J=b)1`ZEFH;lEpRR`aAz+k z#9;#!@rV5N(MCXJc!^eCLX(`*QZrc>Jn|J!*zRM-tZ9~DP&JPYT+>K&f=;|F!>EDA zVtAEv+eR|t=%OaI1sN2I<~`?xX#AneIX~Ktc3E#?s?yz739s}4pCSSg7x5YUbs6W& z47l;w{l*)+wM{;j% zx;?F|l#_8*g82|***e-16-)+N#MP;%>Vdm%5r0a{@yA4!Fd?&;wgM<%NEF{~wb=|w zK7hZG=_0(7%Nw`vgbsJewP7hmVL zjf_-kL8i^@z;qaGN)1rwufuo>~X|59?OYhO1<9Z zwzz&$UCrS%kgVE!;(Ka!8D4v?=`R;djgg!AJlU(W7VU zt{U8F@N;KcXoxDt;Jc*RmS}#ac0pPXqU!1Hsi@KIcUl8@E3?mH3*;Ifv+Ek(gn8l+ zhU}enYr=l1G6aZE$GB+?)P{yLlLKWEREbc6@i0P{NZZ#*y1clx6OTt9qf5|=pxoUA z8K;}s)tYzHHw%V|$cWf8#2pKS?N2k@Yg)esSg(^Ia1A8CE99RdbvNH`k~lsuN1zYw zq>BHLkUVT14Wa}Nu2`GR*$4O*cKSm{mfwBkZ~@sKHo4n*87AM|8|LdkUjE)pO4S!t zy5KR*7G(kb+2xS;t9btWDdS(I0L8;BGoU+s4qz z?WalaPPh%;p#j$y_{P#d8K5#HkY6v0L!V>!BFtlw^?~qQllo}L{yjLDYxK4Y9c9_$ zpLpca<-OI24QlL+BKNH9(YML109-m$j_~u|NfjhFE)riAFZ{6q%+5v<|Ir6KD4T6ip_3Tluz{l}JNe?76seHw>5#;$`J(<=-8`5#MXYt%*C5B({2Cq-L&` zx}Fzu*;xmrvuPKb{}pB1t8I2cF|AAgr%tp^ABa}F^k|du3bxT9952!5&BAY(K{3$! zFJ1o$P8{S#2K1-(UOCeLkfLI`_}-o8KdMkNkK<&9Avn?CDx}&`i;KUkK<-Kv+q7qI z=aq6IQchU^at|7ucumQ0ZQf)PBh>{25UoI}9Z`&7OCVmxr9)5B} zHAei>B}UfM`HHq}chRLO{b6ZmcH|M9*^1`hzut@MQDY-6u(?e7jxb7b@9nlb*ogZx zXdt%@)kSqR-NJQE@aUbl{R1~JKN5W2eD20d%MwKSkDWO2$X8}Nrm@(o2UYTj9yZZe z#GWcjv{%`9tflqkz0$AZW=Dx%*05{Wn#Ro#?2pq&lCj=rcK>WYZxAPrzRSCO>mM=D z$dmT28)BU6cQUcqgpQiQQid2<%RAtBDpgyBmu9{k6F3nN7E}AkL5qt&tMMvwPTH^C-~|OP#sAQ}WqjtrFV7zD zanJ@XMgN}yI*oUpvRKVQNlTY_$GE6K-!$&N5@jH6SCIO?x!gj7;(+XRQ`fnnTq>Bz z09}HgbkdbyTfnd5+eCwi{xK8Y0gpMJ)yzvd{%d6RkG-Qc5|=zB?iZ<;9<=|=gW}WW z_;ENg+d))z=RBw1Llc>JOG%$moNNPrAff+~snKfvmB_?2Wj85S_Ux6AEEM$f>83yR zshC#UDko}YGYC+6wWx)XpfI)B#MIfW%&wf4i5h$E7llNLJkIzct_&D4H4Riqojo#T6zrJuq^i0=77_6Utyh)&~0bGLHxO3 zwT_t*TE#T;_~#sTFh;C5$!mIpDfHtXik|GxEsNN}4e!FPFOC~JF*~AVutCFl{Ep^5 zueUP<8Z+_a@)(Vm!8bwH-<*+1EZs&Q)6=rUu;#_DbIgHxU+#XMqGK_$YBh}aHGOR*Wrs7fn&U}gdZY)={%`!%HNf(NW>I|5d7TA9T9EP#{>{_dUWlmLR@w zIEw~&(21(M_)T5KYZi2id1_h=a%d!QsM1vrQ04T8EYAjwB4GwD4r^NLplA~2vUZNi zq!(X;aYK0qQT&-COpD>gPvIGhzTbwL(}+?*yFjpW$@(7>&E{-R&)vk~#>sd>YWEzM zLC+t=7YDf>PQ3HK+EKCz6!WOL!rBc(Up!Ctpdynj-#!r!TefaP&!GgBP_?yI3`C`s z{g8Nk1}jzEOJcgv3QMIEh4Qk#mVa^kofu2QZg&bJ25LQ{Xw6~=QA9ziv+OOWp_Tu) zAhC<-cyyk!&V}Fhqr6EkX@k%R27`N^TvRy|Q&Y?Fmo}1?pN74!v+#o`t=d6NBvf^A z($7ynpZkue=9R0|!6Q6t@YQuL{E-cAK+-?LwT{|y9 z`5e0L)%#7oC?50%#j89tnobKy@^u6-!YR|S_9r~+3-ya@auz4fI84JBEbSVW5@vht zFIf;*04Py@M4im85A?l%(NRiu&l{%Ew~L{3nnGL+c&b}+IrRftfNvp~DOS#4)+|S? zH_wE*Po9_c8n6@cXM*IP;$Gikicr8j&dPR*kqN9sPgf#GX1uBv8-^X3TGyFJaAzm5 zlnz{Wh8MYN9#@jY>@`YZ%thxfU9)-07f#G_qTY(BVPb#OH1Yt>Ohu_LPs{OGFsVAG z^f)f@vfl2r9_M*pbQfPd=m>xto4MUWBu+ZL1B&~7c5|WcW~I;&Ic$QD-w=ClP|md` zo*strA4W=Ud;2)zs;bs1Pa)M)!rq91Ogn&OHYtfuuTidVxaqueNHPKRs#%1|H{?#Ou#`K*@81uQ&QzuJ}&ktg_VC; z;5Tnj{SIc+GCAT`x^eMwIE!hNQ_ zI_qZn1uK;$)O1(So1}kPq_-Z>HJ@j5DdAj#^jDVnrE4}{{=$j>ubhZGcUiaa*hVmB zoV#_lYU`<`pw{YadS|A3`Z5?$;+BTK+MBk48SrmA?4vPr%Ra@~Z9CL71-D`c8qai9 zD=J6rEWQ=3oIdQiY2hVW$G+0{|bh> zcl1m=Hacjkw!f5=DzOAlS6^&v7dRVA;Kfym?A6peg16a&3-R%zW`;EUg|hf^|L z@Pu}o-7_!kCozH3uZGhUz|%!`xJFB*?|P;#Dkn)d#W5Il2LUh(39YJHH@4(TzS7$z z)53UP|E%skbI=OgqAY=^VP_0F?ikUOCjCp11C6J9ma_PivOWTQ6I-Z2M~@HPXMZk= z$PC1d!-Ledlw}Z!VE!@C-4oT&{PQwpg~ap`D}P+@(<5b7g|{xyU#cM)WToYahEq>< zeXg*m-_5LAGj=t+>VHI;1>czYI=uJxy|P6`FSbdqiP2lqwZK}Fv4c2Et=eT^)kvfP zGvl}OU!!z#SStbhih_maY9~8YL^+-=zpROSjETXX%zyfaTKQW`HskRbr?YYA2G)V{ z%36tpd|dN?`=Cn}Z8D9dwh4j8cC#JFcByM9YPs_Gan&m?4qsc$)1_!tDe?(7ELaOO z?oV)lOy&>n+Q@@G+f_ur3MtRm=2yO=gjT&fRTW3LzE+jZ7&^oG@xv!^UfgxCD5ji~ z&6G`+-2%0_fO$z&UL)wr!T}LJW52cxtw`?O`eF&$^D`E@&sBh|@7}C#bPxmpm-0f8tt1W(Zk~foQe~}?@ts@^eC8-USncWZ3EV%J2_;k zb}aw>VNUhLA@|9YPClzR^i@OV^|g9DT?gu&6aK%A=OF{%d7Sy>QS|Mn?N7_yI+<=A zk6QwPi;*R+V^&bGaD70O{&ebWYDY)p!%t<|y}?88PfVE)#$atcbQ9l-RYU;;)!I#& z$zfC=AoyIOgoK@c>=pke|Pu_*` zh;`0m7Kv|PgR_kG@mCjyq-^wZL*wysc$#${;YR(IT5Mf<@kMGN1wgkKAHyumG(9Bt zge$HNmW{woN}HbE%o!gtnbv@5f()4<7y=o-nqkuouo|}>??Cj})-;f0+4^oi#yUjR ztsBuDnM~RHH*y7Dnl|9gIDuR9_{BwWm0X;!hJsa;gFO01JxGdpO?1SXY)=ls64Zl0 z9}zw^K4BOsd8lxzuYAHIf)f3fS$@$}r%KfPS}jx_S6Rp5zN9>7jSx4ztAo9=Bu)N+ zfDT}blHFDE^G7|#oaI)19nn*i+AVpjK%_VjsqpIuZOO6*eUpx4*#w~pbix&NcOL%U z)<5jXkh0)C^lf}z(3(saobhp)d z)Ddp=lJ9vkk#d7a(- zi`n$B|CjUl>|p%j9@M&wj)cg|+sjgjsc;px0rQ<2cs}jNZyk;Y-vlh;RiJiOJCxla zUWcpk4(O|x?qE0e2wv?044V8TVvJ+|Wd4^5ZcwElA zdM>lnX=LbQs&Ff8^2M9lLxjSFR%Bhl$qFIHV>56IHXau2-MN}jNVL53E1?zM%6i}} zp+xK}Q~jF#E%V)1%}#L$Zr#8Ipi1AZ{%Ft9fbd-!B7Q}z=L7K`soalm*hA2Zbn~cp za7H9%@I4-hJ_%x5bQdg(fo7bg*j>nKDv6UN?97^^J^grlBO#$QNfdy?o^)Z| z8TToT-B~uVH_DLiK}Nf{MU6>@RU4YI&(kOZfR&kFx8R2u=(@XYZ;l-t_LN)=?<^<{ zlI#mFRuKu(Pd}A7Y7L;#(?$V33ik4TXY zs!E6&Pfp|z`gla)FHTZMp;3T0CGLaY3h))c2Br>UaP!*Hf|BHr4SnN7G}DIi3_%r4 zL$s&G{u46zx_EDr*i*0^R+XZQlIFy29e*M17aBh3$P~9fMEH2l3@5GAxuBJb8ryZQrr&QzK73?8>h5D8>T{|)B^k5)RjZAz6?g^ zi#?4k_meBy{GGCcx0bdq$3%t}Kv`syv_3uXmrI<9Ggn6hKh<@uKt5W5^t(Uuyg2h9 zbX~02Q*$NKZ#z4|em^e15ftf4Ud2JsLiU5YaBcRXD#RN6>CJ0%_CNVhsL(h#SQ)l% U_3^mr-{l=S;IKb$@7c@$0VeK(4FCWD diff --git a/docs/articles/08-more-epic_files/header-attrs-2.8/header-attrs.js b/docs/articles/08-more-epic_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/08-more-epic_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/09-convenience-features.html b/docs/articles/09-convenience-features.html index 9c9c7c77..b88f23f9 100644 --- a/docs/articles/09-convenience-features.html +++ b/docs/articles/09-convenience-features.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
                      @@ -39,7 +39,7 @@
                    • @@ -95,7 +98,7 @@
                      • - +
                        • @@ -109,7 +112,7 @@ -
                        +

                        Convenience features

                        @@ -129,30 +132,33 @@

                        Graphical User Interface (GUI) for setting options

                        -

                        I wanted to make {disk.frame} as easy to use as possible. I often forget what options are available to me. So I’ve made a graphical user interface (GUI) which can invoked with

                        -
                        setup_disk.frame(gui = TRUE)
                        -

                        The above opens up a Shiny app where the user can adjust the {disk.frame} options in a .

                        +

                        I wanted to make disk.frame as easy to use as possible. I often forget what options are available to me. So I’ve made a graphical user interface (GUI) which can invoked with

                        +
                        +setup_disk.frame(gui = TRUE)
                        +

                        The above opens up a Shiny app where the user can adjust the disk.frame options in a .

                        RStudio column name completion

                        -
                        library(disk.frame)
-mtcars.df = as.disk.frame(mtcars)
-
-mtcars.df %>% 
-  filter(<press tab>)
                        +
                        library(disk.frame)
+mtcars.df = as.disk.frame(mtcars)
+
+mtcars.df %>% 
+  filter(<press tab>)

                        you can press tab in RStudio and it will show all columns available.

                        Insert ceremony/boilerplate into code in RStudio

                        The below will insert the recommended ceremony code into your editor

                        -
                        disk.frame::insert_ceremony()
                        +
                        +disk.frame::insert_ceremony()

                        should insert

                        -
                        # this willl set disk.frame with multiple workers
-setup_disk.frame()
+
                        +# this willl set disk.frame with multiple workers
+setup_disk.frame()
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
                        
+options(future.globals.maxSize = Inf)

                        @@ -172,7 +178,7 @@

                        -

                        Site built with pkgdown 1.5.1.

                        +

                        Site built with pkgdown 1.6.1.

                        diff --git a/docs/articles/09-convenience-features_files/header-attrs-2.8/header-attrs.js b/docs/articles/09-convenience-features_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/09-convenience-features_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/10-group-by.html b/docs/articles/10-group-by.html index 9c1b79f0..19d19535 100644 --- a/docs/articles/10-group-by.html +++ b/docs/articles/10-group-by.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
                        @@ -39,7 +39,7 @@
                      • @@ -95,7 +98,7 @@
                        • - +
                          • @@ -109,7 +112,7 @@ -
                          +

                          Group-by

                          @@ -134,30 +137,31 @@

                          In the custom one-stage group-by chapter, we will cover

                            -
                          • The architecture of {disk.frame} and its implications for group-by
                            • +
                          • The architecture of disk.frame and its implications for group-by
                          • How to define custom one-stage group-by functions and its limitations

                          One-stage Group-by

                          -

                          A one-stage group-by is the same as group-by for data.frames. This would be unremarkable, if not for the limitations imposed by the disk-based nature of {disk.frame}. Before v0.3.0 of {disk.frame}, one-stage group-by was not possible, and the user had to rely on two-stage group-by even for simple operations like mean.

                          +

                          A one-stage group-by is the same as group-by for data.frames. This would be unremarkable, if not for the limitations imposed by the disk-based nature of disk.frame. Before v0.3.0 of disk.frame, one-stage group-by was not possible, and the user had to rely on two-stage group-by even for simple operations like mean.

                          However, now that one-stage group-by is possible, there are still limitations and not all functions are supported out-of-the-box. Hence, in the next chapter we will describe how to define custom one-stage group-by functions.

                          An example of one-stage group-by:

                          -
                          result_from_disk.frame = iris %>%
-  as.disk.frame %>%
-  group_by(Species) %>%
-  summarize(
-    mean(Petal.Length),
-    sumx = sum(Petal.Length/Sepal.Width),
-    sd(Sepal.Width/ Petal.Length),
-    var(Sepal.Width/ Sepal.Width),
-    l = length(Sepal.Width/ Sepal.Width + 2),
-    max(Sepal.Width),
-    min(Sepal.Width),
-    median(Sepal.Width)
-    ) %>%
-  collect
                          -

                          It is important to note that not all functions that can run in dplyr::summarize would work automatically. This is because of how {disk.frame} works. Please see the next chapter on defining custom one-stage group-by functions to learn more.

                          +
                          +result_from_disk.frame = iris %>% 
+  as.disk.frame %>% 
+  group_by(Species) %>% 
+  summarize(
+    mean(Petal.Length), 
+    sumx = sum(Petal.Length/Sepal.Width), 
+    sd(Sepal.Width/ Petal.Length), 
+    var(Sepal.Width/ Sepal.Width), 
+    l = length(Sepal.Width/ Sepal.Width + 2),
+    max(Sepal.Width), 
+    min(Sepal.Width), 
+    median(Sepal.Width)
+    ) %>% 
+  collect
                          +

                          It is important to note that not all functions that can run in dplyr::summarize would work automatically. This is because of how disk.frame works. Please see the next chapter on defining custom one-stage group-by functions to learn more.

                          List of supported group-by functions

                          @@ -296,69 +300,76 @@

                          Also, notice that there is no guaranteed order for the distribution of the ids to the chunks. The order is random, but each chunk is likely to have a similar number of rows, provided that id does not follow a skewed distribution i.e. where a few distinct values make up the majority of the rows.

                          Typically, chunk_group_by is performed WITHIN each chunk. This is not an issue if the chunks have already been sharded on the by variables beforehand; however, if this is not the case then one may need a second stage aggregation to obtain the correct result, see Two-stage group by.

                          By forcing the user to choose chunk_group_by (within each chunk) and hard_group_by (across all chunks), this ensures that the user is conscious of the choice they are making. In sparklyr the equivalent of a hard_group_by is performed, which we should avoid, where possible, as it is time-consuming and expensive. Hence, disk.frame has chosen to explain the theory and allow the user to make a conscious choice when performing group_by.

                          -
                          suppressMessages(library(disk.frame))
-flights.df %>%
-  hard_group_by(carrier) %>% # notice that hard_group_by needs to be set
-  chunk_summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
-  collect %>%
-  arrange(carrier)
                          +
                          +suppressMessages(library(disk.frame))
+flights.df %>%
+  hard_group_by(carrier) %>% # notice that hard_group_by needs to be set
+  chunk_summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
+  collect %>% 
+  arrange(carrier)

                          Two-Stage Group-by

                          -

                          Prior to {disk.frame} v0.3.0, there is no general support for one-stage group-by. Hence a two-stage style group-by is needed. The key is understand is the chunk_group_by which performs group-by within each chunk.

                          +

                          Prior to disk.frame v0.3.0, there is no general support for one-stage group-by. Hence a two-stage style group-by is needed. The key is understand is the chunk_group_by which performs group-by within each chunk.

                          For most group-by tasks, the user can achieve the desired result WITHOUT using hard = TRUE by performing the group by in two stages. For example, suppose you aim to count the number of rows group by carrier, you can set hard = F to find the count within each chunk and then use a second group-by to summaries each chunk’s results into the desired result. For example,

                          -
                          flights.df %>%
-  chunk_group_by(carrier) %>% # `chunk_group_by` aggregates within each chunk
-  chunk_summarize(count = n()) %>%  # mean follows normal R rules
-  collect %>%  # collect each individul chunks results and row-bind into a data.table
-  group_by(carrier) %>%
-  summarize(count = sum(count)) %>%
-  arrange(carrier)
                          +
                          +flights.df %>%
+  chunk_group_by(carrier) %>% # `chunk_group_by` aggregates within each chunk
+  chunk_summarize(count = n()) %>%  # mean follows normal R rules
+  collect %>%  # collect each individul chunks results and row-bind into a data.table
+  group_by(carrier) %>% 
+  summarize(count = sum(count)) %>% 
+  arrange(carrier)

                          Because this two-stage approach avoids the expensive hard group_by operation, it is often significantly faster. However, it can be tedious to write; and this is a con of the disk.frame chunking mechanism.

                          Note: this two-stage approach is similar to a map-reduce operation.

                          - -
                          flights.df = as.disk.frame(nycflights13::flights)
+
+
                          +flights.df = as.disk.frame(nycflights13::flights)
 
-flights.df %>%
-  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
-  chunk_group_by(year) %>%
-  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
-  collect
                          
+flights.df %>%
+  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
+  chunk_group_by(year) %>% 
+  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
+  collect

                          This is two-stage group-by in action

                          -
                          # need a 2nd stage to finalise summing
-flights.df %>%
-  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
-  chunk_group_by(year) %>%
-  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
-  collect %>%
-  group_by(year) %>%
-  summarise(sum_dist = sum(sum_dist))
                          +
                          +# need a 2nd stage to finalise summing
+flights.df %>%
+  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
+  chunk_group_by(year) %>% 
+  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
+  collect %>% 
+  group_by(year) %>% 
+  summarise(sum_dist = sum(sum_dist))

                          You can mix group-by with other dplyr verbs as below, here is an example of using filter.

                          -
                          # filter
-pt = proc.time()
-df_filtered <-
-  flights.df %>%
-  filter(month == 1)
-cat("filtering a < 0.1 took: ", data.table::timetaken(pt), "\n")
-nrow(df_filtered)
                          +
                          +# filter
+pt = proc.time()
+df_filtered <-
+  flights.df %>% 
+  filter(month == 1)
+cat("filtering a < 0.1 took: ", data.table::timetaken(pt), "\n")
+nrow(df_filtered)

                          Hard group-by

                          Another way to perform a one-stage group_by is to perform a hard_group_by on a disk.frame. This will rechunk the disk.frame by the by-columns. This is not recommended for performance reasons, as it can be quite slow to rechunk the file chunks on disk.

                          -
                          pt = proc.time()
-res1 <- flights.df %>%
-  srckeep(c("month", "dep_delay")) %>%
-  filter(month <= 6) %>%
-  mutate(qtr = ifelse(month <= 3, "Q1", "Q2")) %>%
-  hard_group_by(qtr) %>% # hard group_by is MUCH SLOWER but avoid a 2nd stage aggregation
-  chunk_summarise(avg_delay = mean(dep_delay, na.rm = TRUE)) %>%
-  collect
-cat("group-by took: ", data.table::timetaken(pt), "\n")
+
                          +pt = proc.time()
+res1 <- flights.df %>% 
+  srckeep(c("month", "dep_delay")) %>% 
+  filter(month <= 6) %>% 
+  mutate(qtr = ifelse(month <= 3, "Q1", "Q2")) %>% 
+  hard_group_by(qtr) %>% # hard group_by is MUCH SLOWER but avoid a 2nd stage aggregation
+  chunk_summarise(avg_delay = mean(dep_delay, na.rm = TRUE)) %>% 
+  collect
+cat("group-by took: ", data.table::timetaken(pt), "\n")
 
-collect(res1)
                          
+collect(res1)

                          @@ -378,7 +389,7 @@

                          -

                          Site built with pkgdown 1.5.1.

                          +

                          Site built with pkgdown 1.6.1.

                          diff --git a/docs/articles/10-group-by_files/header-attrs-2.8/header-attrs.js b/docs/articles/10-group-by_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/10-group-by_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/11-custom-group-by.html b/docs/articles/11-custom-group-by.html index 4d0952c0..53eb0bf7 100644 --- a/docs/articles/11-custom-group-by.html +++ b/docs/articles/11-custom-group-by.html @@ -31,7 +31,7 @@ disk.frame - 0.3.6 + 0.5.0
                          @@ -39,7 +39,7 @@
                        • @@ -95,7 +98,7 @@
                          • - +
                            • @@ -109,7 +112,7 @@ -
                            +

                            Custom One-Stage Group-by functions

                            @@ -127,27 +130,28 @@

                            ZJ

                            Custom one-stage group-by

                            Group-by is one the most useful and commonly used operations in data manipulation. The most common to perform group with dplyr syntax would be something similar to the below

                            -
                            dataf %>%
-  group_by(grpA) %>%
-  summarize(mean(X))
                            -

                            However, the above nice and concise group-by syntax creates a challenge in {disk.frame}, because {disk.frame} organises the data in chunks and these chunks are often manipulated independently to each other which makes it difficult to compute the mean of any column if the group-by column is stored in different chunks. In this chapter, I will explain how {disk.frame} achieve the one-stage group-by syntax that you see, and how you can define custom one-stage group-by functions for use with {disk.frame}.

                            +
                            +dataf %>% 
+  group_by(grpA) %>% 
+  summarize(mean(X))
                            +

                            However, the above nice and concise group-by syntax creates a challenge in disk.frame, because disk.frame organises the data in chunks and these chunks are often manipulated independently to each other which makes it difficult to compute the mean of any column if the group-by column is stored in different chunks. In this chapter, I will explain how disk.frame achieve the one-stage group-by syntax that you see, and how you can define custom one-stage group-by functions for use with disk.frame.

                            At a glance

                            -

                            {disk.frame} allows the user to create custom one-stage group-by functions. To make a function fn one stage, one needs to define two functions:

                            +

                            disk.frame allows the user to create custom one-stage group-by functions. To make a function fn one stage, one needs to define two functions:

                            1. fn_df.chunk_agg.disk.frame which applies itself to each chunk
                            2. fn_df.collected_agg.disk.frame which accepts a list of returns from fn_df.chunk_agg.disk.frame and finalize the computation
                            -

                            For example, to make mean a one-stage group-by function, {disk.frame} has defined mean_df.chunk_agg.disk.frame and mean_df.collected_agg.disk.frame, which we will illustrate with examples below.

                            -

                            But first, we shall explain some theory behind {disk.frame} to help you better understand “why does {disk.frame} do it like that?”.

                            +

                            For example, to make mean a one-stage group-by function, disk.frame has defined mean_df.chunk_agg.disk.frame and mean_df.collected_agg.disk.frame, which we will illustrate with examples below.

                            +

                            But first, we shall explain some theory behind disk.frame to help you better understand “why does disk.frame do it like that?”.

                            How does {disk.frame} work

                            -

                            One may ask, how come only a few functions are supported for one-stage group-by? And why are some functions like median only produce estimates instead of producing the exact figure? To answer these questions, we need to have an understanding of how {disk.frame} works.

                            +

                            One may ask, how come only a few functions are supported for one-stage group-by? And why are some functions like median only produce estimates instead of producing the exact figure? To answer these questions, we need to have an understanding of how disk.frame works.

                            A disk.frame is organized as chunks stored on disk. Each chunk is a file stored in fst format. The {future} package is used to apply the same function to each chunk, each of these operations are carried out in a separate R session. These R sessions cannot communicate with each other during the execution of the operations.

                            Once the operations have been performed on all chunks, the results will be bought back to the session from which the operations were called. This is the only point of inter-process communication.

                            To summarize, the two phases of a df %>% some_fn %>% collect operation are

                            @@ -160,60 +164,65 @@

                            How group-by works

                            -

                            Except for passing the result back to the main session, communication between worker sessions are not allowed. This limits how group-by operations can be performed, hence why group-by was done in two stages prior to {disk.frame} v0.3.0. However, R’s meta-programming abilities allow us to rewrite code to perform the two-stage group-bys using one-stage group-by code. For example, consider:

                            -
                            df %>%
-  group_by(grp1) %>%
-  summarize(sum(x)) %>%
-  collect
                            +

                            Except for passing the result back to the main session, communication between worker sessions are not allowed. This limits how group-by operations can be performed, hence why group-by was done in two stages prior to disk.frame v0.3.0. However, R’s meta-programming abilities allow us to rewrite code to perform the two-stage group-bys using one-stage group-by code. For example, consider:

                            +
                            +df %>% 
+  group_by(grp1) %>% 
+  summarize(sum(x)) %>% 
+  collect

                            we can use meta-programming to transform that to

                            -
                            df %>% 
-  chunk_group_by(grp1) %>% 
-  chunk_summarize(__tmp1__= sum(x)) %>% 
-  collect() %>% 
-  group_by(grp1) %>% 
-  summarize(x = sum(__tmp1__))
                            +
                            df %>% 
+  chunk_group_by(grp1) %>% 
+  chunk_summarize(__tmp1__= sum(x)) %>% 
+  collect() %>% 
+  group_by(grp1) %>% 
+  summarize(x = sum(__tmp1__))

                            Basically, we are “compiling” one-stage group-by code to two-stage group-by code, and then executing it.

                            For mean, it’s trickier, as one needs to keep track on the numerator and the denominator separately in computing mean(x) = sum(x)/length(x).

                            -

                            Therefore, {disk.frame} compiles

                            -
                            df %>%
-  group_by(grp1) %>%
-  summarize(meanx = mean(x)) %>%
-  collect
                            +

                            Therefore, disk.frame compiles

                            +
                            +df %>% 
+  group_by(grp1) %>% 
+  summarize(meanx = mean(x)) %>% 
+  collect

                            to

                            -
                            df %>% 
-  chunk_group_by(grp1) %>% 
-  chunk_summarize(__tmp1__ = list(mean_df.chunk_agg.disk.frame(x))) %>% 
-  collect %>% 
-  group_by(grp1) %>% 
-  chunk_summarize(meanx = mean_df.chunk_agg.disk.frame(__tmp1__))
                            +
                            df %>% 
+  chunk_group_by(grp1) %>% 
+  chunk_summarize(__tmp1__ = list(mean_df.chunk_agg.disk.frame(x))) %>% 
+  collect %>% 
+  group_by(grp1) %>% 
+  chunk_summarize(meanx = mean_df.chunk_agg.disk.frame(__tmp1__))

                            where mean_df.chunk_agg.disk.frame defines what needs to be done to each chunk, as you can see, the return value is a vector where the elements are named sumx and lengthx. Also note because the return value is not a scalar, we need to write it in a list (line 3).

                            Here is an example implementation of mean.chunk_agg.disk.frame

                            -
                            mean_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE, ...) {
-  sumx = sum(x, na.rm = na.rm)
-  lengthx = length(x) - ifelse(na.rm, sum(is.na(x)), 0)
-  c(sumx = sumx, lengthx = lengthx)
-}
                            +
                            +mean_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE, ...) {
+  sumx = sum(x, na.rm = na.rm)
+  lengthx = length(x) - ifelse(na.rm, sum(is.na(x)), 0)
+  c(sumx = sumx, lengthx = lengthx)
+}

                            The mean_df.collected_agg.disk.frame receives a list of outputs from mean_df.chunk_agg.disk.frame. Recall that mean.chunk_agg.disk.frame returns a vector for each chunk, so the input to mean_df.collected_agg.disk.frame is a list of vectors

                            -
                            mean_df.collected_agg.disk.frame <- function(listx) {
-  sum(sapply(listx, function(x) x["sumx"]))/sum(sapply(listx, function(x) x["lengthx"]))
-}
                            +
                            +mean_df.collected_agg.disk.frame <- function(listx) {
+  sum(sapply(listx, function(x) x["sumx"]))/sum(sapply(listx, function(x) x["lengthx"]))
+}

                            How to define custom one-stage group-by functions

                            Now that we have seen two examples, namely sum and mean, we are ready to summarize how group-by functions are implemented.

                            Given the below

                            -
                            df %>%
-  group_by(grp1) %>%
-  summarize(namex = fn(x)) %>%
-  collect
                            -

                            {disk.frame} compiles it to

                            -
                            df %>% 
-  chunk_group_by(grp1) %>% 
-  chunk_summarize(__tmp1__ = list(fn_df.chunk_agg.disk.frame(x))) %>% 
-  collect %>% 
-  group_by(grp1) %>% 
-  chunk_summarize(namex = fn_df.collected_agg.disk.frame(__tmp1__))
                            +
                            +df %>% 
+  group_by(grp1) %>% 
+  summarize(namex = fn(x)) %>% 
+  collect
                            +

                            disk.frame compiles it to

                            +
                            df %>% 
+  chunk_group_by(grp1) %>% 
+  chunk_summarize(__tmp1__ = list(fn_df.chunk_agg.disk.frame(x))) %>% 
+  collect %>% 
+  group_by(grp1) %>% 
+  chunk_summarize(namex = fn_df.collected_agg.disk.frame(__tmp1__))

                            Based on the above information, to make fn a one-stage group-by function, the user has to

                            1. Define fn_df.chunk_agg.disk.frame which is a function to be applied to each chunk
                              2. @@ -225,38 +234,42 @@

                            2. Define sum_df.chunk_agg.disk.frame

                            -
                            sum_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE) {
-  sum(x, na.rm=na.rm)
-}
                            +
                            +sum_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE) {
+  sum(x, na.rm=na.rm)
+}
                            1. Define sum_df.collected_agg.disk.frame, which needs to accept a list of sum(x, na.rm), but sum(x, na.rm) is just a numeric, so
                            -
                            sum_df.collected_agg.disk.frame <- function(list_sum) {
-  sum(unlist(list_sum))
-}
                            +
                            +sum_df.collected_agg.disk.frame <- function(list_sum) {
+  sum(unlist(list_sum))
+}

                            Example of implementing n_distinct:

                            The dplyr::n_distinct function counts the number of distint values from a vector x

                            1. Define n_distinct_df.chunk_agg.disk.frame, to return a list of unique values. Because the same value can appear in multiple chunks, to ensure that we don’t double count, we simply return all the unique values from each chunk which is then de-duplicated in the next phase
                            -
                            n_distinct_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE) {
-  if(na.rm) {
-    setdiff(unique(x), NA)
-  } else {
-    unique(x)
-  }
-}
                            +
                            +n_distinct_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE) {
+  if(na.rm) {
+    setdiff(unique(x), NA)
+  } else {
+    unique(x)
+  }
+}
                            1. Define n_distinct_df.collected_agg.disk.frame, which de-duplicates the unique values
                            -
                            n_distinct_df.collected_agg.disk.frame <- function(list_of_chunkwise_uniques) {
-  dplyr::n_distinct(unlist(list_of_chunkwise_uniques))
-}
                            +
                            +n_distinct_df.collected_agg.disk.frame <- function(list_of_chunkwise_uniques) {
+  dplyr::n_distinct(unlist(list_of_chunkwise_uniques))
+}

                            Limitations

                            -

                            We have seen that {disk.frame} performs operations in two phases

                            +

                            We have seen that disk.frame performs operations in two phases

                            1. apply the same function to each chunk
                            2. row-bind the results
                              3. @@ -283,7 +296,7 @@

                            -

                            Site built with pkgdown 1.5.1.

                            +

                            Site built with pkgdown 1.6.1.

                            diff --git a/docs/articles/11-custom-group-by_files/header-attrs-2.8/header-attrs.js b/docs/articles/11-custom-group-by_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/11-custom-group-by_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/88-trouble-shooting.html b/docs/articles/88-trouble-shooting.html new file mode 100644 index 00000000..336e9c5b --- /dev/null +++ b/docs/articles/88-trouble-shooting.html @@ -0,0 +1,186 @@ + + + + + + + +Trouble shooting • disk.frame + + + + + + + + + + +
                            +
                            + + + + +
                            +
                            +
                            +

                            Trouble shooting

                            +

                            ZJ

                            + + + Source: vignettes/88-trouble-shooting.Rmd + + +
                            + + + +
                            +

                            +Steps to trouble shoot

                            +
                              +

                            1. I suggest updating {future} and your R version if you have not already done so.

                              2. +

                            2. Are you able to share the data?

                              3. +

                            3. Do a good MWE

                              4. +
                            +
                            library(disk.frame)
+setup_disk.frame()
+
+df<-as.disk.frame(a)
+
+
+df1 = mutate(df, date = as.Date(as.character(datadate), format="%Y%m%d"))
+
+head(df1)
                            +
                              +

                            1. Check if your virus scanner is blocking interprocess communication

                              2. +

                            2. Try to apply the function to just one chunk, perhaps there is a syntax error or column error? If one chunk works then you can rule out coding error

                              3. +
                            +
                            get_chunk(df, 1) %>%
+  mutate(date = as.Date(as.character(datadate), format="%Y%m%d"))
                            +
                              +
                            1. Set the number of workers to 1, so there is no more inter-process communication. Does it work now? If it does, then it’s the inter process communication. You might need to contact your admin for help
                              2. +
                            +
                            setup_disk.frame(workers=1)
+mutate(df, date = as.Date(as.character(datadate), format="%Y%m%d"))
+As an MWE this works for me.
+
+a = data.frame(datadate = rep("20201007", 3e6))
                            +
                            +
                            + + + +
                            + + + +
                            +

                            Developed by Dai ZJ.

                            +
                            + +
                            +

                            Site built with pkgdown 1.6.1.

                            +
                            + +
                            +
                            + + + + + + diff --git a/docs/articles/88-trouble-shooting_files/header-attrs-2.8/header-attrs.js b/docs/articles/88-trouble-shooting_files/header-attrs-2.8/header-attrs.js new file mode 100644 index 00000000..dd57d92e --- /dev/null +++ b/docs/articles/88-trouble-shooting_files/header-attrs-2.8/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/index.html b/docs/articles/index.html index 7e29886b..3569fe74 100644 --- a/docs/articles/index.html +++ b/docs/articles/index.html @@ -71,7 +71,7 @@ disk.frame - 0.3.6 + 0.5.0

                            @@ -79,7 +79,7 @@
                          • @@ -135,7 +138,7 @@
                            • @@ -194,7 +199,7 @@

                            All vignettes

                            -

                            Site built with pkgdown 1.5.1.

                            +

                            Site built with pkgdown 1.6.1.

                            diff --git a/docs/authors.html b/docs/authors.html index fd9e2f46..a2bdd37f 100644 --- a/docs/authors.html +++ b/docs/authors.html @@ -71,7 +71,7 @@ disk.frame - 0.3.6 + 0.5.0
                            @@ -79,7 +79,7 @@
                          • @@ -135,7 +138,7 @@
                            • - +
                              • @@ -178,7 +181,7 @@

                              Authors

                          • -

                          Site built with pkgdown 1.5.1.

                          +

                          Site built with pkgdown 1.6.1.

                          diff --git a/docs/index.html b/docs/index.html index 41066c9b..d63f78e2 100644 --- a/docs/index.html +++ b/docs/index.html @@ -35,7 +35,7 @@ disk.frame - 0.3.6 + 0.5.0
                      • @@ -43,7 +43,7 @@
                    • @@ -99,7 +102,7 @@
                      • - +
                        • @@ -125,25 +128,31 @@

                        Introduction

                        How do I manipulate tabular data that doesn’t fit into Random Access Memory (RAM)?

                        -

                        Use {disk.frame}!

                        -

                        In a nutshell, {disk.frame} makes use of two simple ideas

                        -

                        1) split up a larger-than-RAM dataset into chunks and store each chunk in a separate file inside a folder and 2) provide a convenient API to manipulate these chunks

                        -

                        {disk.frame} performs a similar role to distributed systems such as Apache Spark, Python’s Dask, and Julia’s JuliaDB.jl for medium data which are datasets that are too large for RAM but not quite large enough to qualify as big data.

                        +

                        Use disk.frame!

                        +

                        In a nutshell, disk.frame makes use of two simple ideas

                        +
                          +
                        1. split up a larger-than-RAM dataset into chunks and store each chunk in a separate file inside a folder and
                          2. +
                        2. provide a convenient API to manipulate these chunks
                          3. +
                        +

                        disk.frame performs a similar role to distributed systems such as Apache Spark, Python’s Dask, and Julia’s JuliaDB.jl for medium data which are datasets that are too large for RAM but not quite large enough to qualify as big data.

                        Installation

                        -

                        You can install the released version of {disk.frame} from CRAN with:

                        -
                        install.packages("disk.frame")
                        +

                        You can install the released version of disk.frame from CRAN with:

                        +
                        +install.packages("disk.frame")

                        And the development version from GitHub with:

                        -
                        # install.packages("devtools")
-devtools::install_github("xiaodaigh/disk.frame")
                        +
                        +# install.packages("devtools")
+devtools::install_github("xiaodaigh/disk.frame")

                        On some platforms, such as SageMaker, you may need to explicitly specify a repo like this

                        -
                        install.packages("disk.frame", repo="https://cran.rstudio.com")
                        +
                        +install.packages("disk.frame", repo="https://cran.rstudio.com")

                        Vignettes and articles

                        -

                        Please see these vignettes and articles about {disk.frame}

                        +

                        Please see these vignettes and articles about disk.frame

                        • Quick start: {disk.frame} which replicates the sparklyr vignette for manipulating the nycflights13 flights data.
                          • @@ -169,33 +178,33 @@

                          a) What is {disk.frame} and why create it?

                          -

                          {disk.frame} is an R package that provides a framework for manipulating larger-than-RAM structured tabular data on disk efficiently. The reason one would want to manipulate data on disk is that it allows arbitrarily large datasets to be processed by R. In other words, we go from “R can only deal with data that fits in RAM” to “R can deal with any data that fits on disk”. See the next section.

                          +

                          disk.frame is an R package that provides a framework for manipulating larger-than-RAM structured tabular data on disk efficiently. The reason one would want to manipulate data on disk is that it allows arbitrarily large datasets to be processed by R. In other words, we go from “R can only deal with data that fits in RAM” to “R can deal with any data that fits on disk”. See the next section.

                          b) How is it different to data.frame and data.table?

                          A data.frame in R is an in-memory data structure, which means that R must load the data in its entirety into RAM. A corollary of this is that only data that can fit into RAM can be processed using data.frames. This places significant restrictions on what R can process with minimal hassle.

                          -

                          In contrast, {disk.frame} provides a framework to store and manipulate data on the hard drive. It does this by loading only a small part of the data, called a chunk, into RAM; process the chunk, write out the results and repeat with the next chunk. This chunking strategy is widely applied in other packages to enable processing large amounts of data in R, for example, see chunkded arkdb, and iotools.

                          -

                          Furthermore, there is a row-limit of 2^31 for data.frames in R; hence an alternate approach is needed to apply R to these large datasets. The chunking mechanism in {disk.frame} provides such an avenue to enable data manipulation beyond the 2^31 row limit.

                          +

                          In contrast, disk.frame provides a framework to store and manipulate data on the hard drive. It does this by loading only a small part of the data, called a chunk, into RAM; process the chunk, write out the results and repeat with the next chunk. This chunking strategy is widely applied in other packages to enable processing large amounts of data in R, for example, see chunkded arkdb, and iotools.

                          +

                          Furthermore, there is a row-limit of 2^31 for data.frames in R; hence an alternate approach is needed to apply R to these large datasets. The chunking mechanism in disk.frame provides such an avenue to enable data manipulation beyond the 2^31 row limit.

                          c) How is {disk.frame} different to previous “big” data solutions for R?

                          -

                          R has many packages that can deal with larger-than-RAM datasets, including ff and bigmemory. However, ff and bigmemory restrict the user to primitive data types such as double, which means they do not support character (string) and factor types. In contrast, {disk.frame} makes use of data.table::data.table and data.frame directly, so all data types are supported. Also, {disk.frame} strives to provide an API that is as similar to data.frame’s where possible. {disk.frame} supports many dplyr verbs for manipulating disk.frames.

                          -

                          Additionally, {disk.frame} supports parallel data operations using infrastructures provided by the excellent future package to take advantage of multi-core CPUs. Further, {disk.frame} uses state-of-the-art data storage techniques such as fast data compression, and random access to rows and columns provided by the fst package to provide superior data manipulation speeds.

                          +

                          R has many packages that can deal with larger-than-RAM datasets, including ff and bigmemory. However, ff and bigmemory restrict the user to primitive data types such as double, which means they do not support character (string) and factor types. In contrast, disk.frame makes use of data.table::data.table and data.frame directly, so all data types are supported. Also, disk.frame strives to provide an API that is as similar to data.frame’s where possible. disk.frame supports many dplyr verbs for manipulating disk.frames.

                          +

                          Additionally, disk.frame supports parallel data operations using infrastructures provided by the excellent future package to take advantage of multi-core CPUs. Further, disk.frame uses state-of-the-art data storage techniques such as fast data compression, and random access to rows and columns provided by the fst package to provide superior data manipulation speeds.

                          d) How does {disk.frame} work?

                          -

                          {disk.frame} works by breaking large datasets into smaller individual chunks and storing the chunks in fst files inside a folder. Each chunk is a fst file containing a data.frame/data.table. One can construct the original large dataset by loading all the chunks into RAM and row-bind all the chunks into one large data.frame. Of course, in practice this isn’t always possible; hence why we store them as smaller individual chunks.

                          -

                          {disk.frame} makes it easy to manipulate the underlying chunks by implementing dplyr functions/verbs and other convenient functions (e.g. the cmap(a.disk.frame, fn, lazy = F) function which applies the function fn to each chunk of a.disk.frame in parallel). So that {disk.frame} can be manipulated in a similar fashion to in-memory data.frames.

                          +

                          disk.frame works by breaking large datasets into smaller individual chunks and storing the chunks in fst files inside a folder. Each chunk is a fst file containing a data.frame/data.table. One can construct the original large dataset by loading all the chunks into RAM and row-bind all the chunks into one large data.frame. Of course, in practice this isn’t always possible; hence why we store them as smaller individual chunks.

                          +

                          disk.frame makes it easy to manipulate the underlying chunks by implementing dplyr functions/verbs and other convenient functions (e.g. the cmap(a.disk.frame, fn, lazy = F) function which applies the function fn to each chunk of a.disk.frame in parallel). So that disk.frame can be manipulated in a similar fashion to in-memory data.frames.

                          e) How is {disk.frame} different from Spark, Dask, and JuliaDB.jl?

                          -

                          Spark is primarily a distributed system that also works on a single machine. Dask is a Python package that is most similar to {disk.frame}, and JuliaDB.jl is a Julia package. All three can distribute work over a cluster of computers. However, {disk.frame} currently cannot distribute data processes over many computers, and is, therefore, single machine focused.

                          -

                          In R, one can access Spark via sparklyr, but that requires a Spark cluster to be set up. On the other hand {disk.frame} requires zero-setup apart from running install.packages("disk.frame") or devtools::install_github("xiaodaigh/disk.frame").

                          -

                          Finally, Spark can only apply functions that are implemented for Spark, whereas {disk.frame} can use any function in R including user-defined functions.

                          +

                          Spark is primarily a distributed system that also works on a single machine. Dask is a Python package that is most similar to disk.frame, and JuliaDB.jl is a Julia package. All three can distribute work over a cluster of computers. However, disk.frame currently cannot distribute data processes over many computers, and is, therefore, single machine focused.

                          +

                          In R, one can access Spark via sparklyr, but that requires a Spark cluster to be set up. On the other hand disk.frame requires zero-setup apart from running install.packages("disk.frame") or devtools::install_github("xiaodaigh/disk.frame").

                          +

                          Finally, Spark can only apply functions that are implemented for Spark, whereas disk.frame can use any function in R including user-defined functions.

                      • @@ -206,29 +215,31 @@

                      Set-up {disk.frame}

                      -

                      {disk.frame} works best if it can process multiple data chunks in parallel. The best way to set-up {disk.frame} so that each CPU core runs a background worker is by using

                      -
                      setup_disk.frame()
+
                      disk.frame works best if it can process multiple data chunks in parallel. The best way to set-up disk.frame so that each CPU core runs a background worker is by using
                      
+
                      +setup_disk.frame()
 
 # this allows large datasets to be transferred between sessions
-options(future.globals.maxSize = Inf)
                      
+options(future.globals.maxSize = Inf)

                      The setup_disk.frame() sets up background workers equal to the number of CPU cores; please note that, by default, hyper-threaded cores are counted as one not two.

                      Alternatively, one may specify the number of workers using setup_disk.frame(workers = n).

                    • Quick-start

                    -
                    suppressPackageStartupMessages(library(disk.frame))
-library(nycflights13)
+
                    +suppressPackageStartupMessages(library(disk.frame))
+library(nycflights13)
 
 # this will setup disk.frame's parallel backend with number of workers equal to the number of CPU cores (hyper-threaded cores are counted as one not two)
-setup_disk.frame()
+setup_disk.frame()
 #> The number of workers available for disk.frame is 6
 # this allows large datasets to be transferred between sessions
-options(future.globals.maxSize = Inf)
+options(future.globals.maxSize = Inf)
 
 # convert the flights data.frame to a disk.frame
 # optionally, you may specify an outdir, otherwise, the 
-flights.df <- as.disk.frame(nycflights13::flights)
                    
+flights.df <- as.disk.frame(nycflights13::flights)

                    @@ -237,35 +248,37 @@

                    dplyr verbs

                    {disk.frame} aims to support as many dplyr verbs as possible. For example

                    -
                    flights.df %>%
-  filter(year == 2013) %>%
-  mutate(origin_dest = paste0(origin, dest)) %>%
-  head(2)
-#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight tailnum origin
-#> 1 2013     1   1      517            515         2      830            819        11      UA   1545  N14228    EWR
-#> 2 2013     1   1      533            529         4      850            830        20      UA   1714  N24211    LGA
-#>   dest air_time distance hour minute           time_hour origin_dest
-#> 1  IAH      227     1400    5     15 2013-01-01 05:00:00      EWRIAH
-#> 2  IAH      227     1416    5     29 2013-01-01 05:00:00      LGAIAH
                    +
                    +flights.df %>% 
+  filter(year == 2013) %>% 
+  mutate(origin_dest = paste0(origin, dest)) %>% 
+  head(2)
+#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight
+#> 1 2013     1   1      517            515         2      830            819        11      UA   1545
+#> 2 2013     1   1      533            529         4      850            830        20      UA   1714
+#>   tailnum origin dest air_time distance hour minute           time_hour origin_dest
+#> 1  N14228    EWR  IAH      227     1400    5     15 2013-01-01 05:00:00      EWRIAH
+#> 2  N24211    LGA  IAH      227     1416    5     29 2013-01-01 05:00:00      LGAIAH

                    Group-by

                    -

                    Starting from {disk.frame} v0.3.0, there is group_by support for a limited set of functions. For example:

                    -
                    result_from_disk.frame = iris %>%
-  as.disk.frame %>%
-  group_by(Species) %>%
-  summarize(
-    mean(Petal.Length),
-    sumx = sum(Petal.Length/Sepal.Width),
-    sd(Sepal.Width/ Petal.Length),
-    var(Sepal.Width/ Sepal.Width),
-    l = length(Sepal.Width/ Sepal.Width + 2),
-    max(Sepal.Width),
-    min(Sepal.Width),
-    median(Sepal.Width)
-    ) %>%
-  collect
                    +

                    Starting from disk.frame v0.3.0, there is group_by support for a limited set of functions. For example:

                    +
                    +result_from_disk.frame = iris %>% 
+  as.disk.frame %>% 
+  group_by(Species) %>% 
+  summarize(
+    mean(Petal.Length), 
+    sumx = sum(Petal.Length/Sepal.Width), 
+    sd(Sepal.Width/ Petal.Length), 
+    var(Sepal.Width/ Sepal.Width), 
+    l = length(Sepal.Width/ Sepal.Width + 2),
+    max(Sepal.Width), 
+    min(Sepal.Width), 
+    median(Sepal.Width)
+    ) %>% 
+  collect

                    The results should be exactly the same as if applying the same group-by operations on a data.frame. If not, please report a bug.

                    @@ -358,69 +371,67 @@

                    Example: data.table syntax

                    -
                    library(data.table)
-#> data.table 1.12.8 using 6 threads (see ?getDTthreads).  Latest news: r-datatable.com
-#> 
-#> Attaching package: 'data.table'
-#> The following object is masked from 'package:purrr':
-#> 
-#>     transpose
-#> The following objects are masked from 'package:dplyr':
-#> 
-#>     between, first, last
+
                    +library(data.table)
 
-suppressWarnings(
-  grp_by_stage1 <-
-    flights.df[
-      keep = c("month", "distance"), # this analysis only required "month" and "dist" so only load those
-      month <= 6,
-      .(sum_dist = sum(distance)),
-      .(qtr = ifelse(month <= 3, "Q1", "Q2"))
-      ]
-)
+suppressWarnings(
+  grp_by_stage1 <- 
+    flights.df[
+      keep = c("month", "distance"), # this analysis only required "month" and "dist" so only load those
+      month <= 6, 
+      .(sum_dist = sum(distance)), 
+      .(qtr = ifelse(month <= 3, "Q1", "Q2"))
+      ]
+)
 
-grp_by_stage1
+grp_by_stage1
 #>    qtr sum_dist
 #> 1:  Q1 27188805
 #> 2:  Q1   953578
 #> 3:  Q1 53201567
 #> 4:  Q2  3383527
 #> 5:  Q2 58476357
-#> 6:  Q2 27397926
                    
+#> 6:  Q2 27397926

                    The result grp_by_stage1 is a data.table so we can finish off the two-stage aggregation using data.table syntax

                    -
                    grp_by_stage2 = grp_by_stage1[,.(sum_dist = sum(sum_dist)), qtr]
+
                    +grp_by_stage2 = grp_by_stage1[,.(sum_dist = sum(sum_dist)), qtr]
 
-grp_by_stage2
+grp_by_stage2
 #>    qtr sum_dist
 #> 1:  Q1 81343950
-#> 2:  Q2 89257810
                    
+#> 2:  Q2 89257810

                    Basic info

                    To find out where the disk.frame is stored on disk:

                    -
                    # where is the disk.frame stored
-attr(flights.df, "path")
-#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpSyWIOI\\file26ac7d293556.df"
                    +
                    +# where is the disk.frame stored
+attr(flights.df, "path")
+#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpEdhlRv\\file33ec607f12cc.df"

                    A number of data.frame functions are implemented for disk.frame

                    -
                    # get first few rows
-head(flights.df, 1)
-#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight tailnum origin
-#> 1: 2013     1   1      517            515         2      830            819        11      UA   1545  N14228    EWR
-#>    dest air_time distance hour minute           time_hour
-#> 1:  IAH      227     1400    5     15 2013-01-01 05:00:00
                    -
                    # get last few rows
-tail(flights.df, 1)
-#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight tailnum origin
-#> 1: 2013     9  30       NA            840        NA       NA           1020        NA      MQ   3531  N839MQ    LGA
-#>    dest air_time distance hour minute           time_hour
-#> 1:  RDU       NA      431    8     40 2013-09-30 08:00:00
                    -
                    # number of rows
-nrow(flights.df)
-#> [1] 336776
                    -
                    # number of columns
-ncol(flights.df)
-#> [1] 19
                    +
                    +# get first few rows
+head(flights.df, 1)
+#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier
+#> 1: 2013     1   1      517            515         2      830            819        11      UA
+#>    flight tailnum origin dest air_time distance hour minute           time_hour
+#> 1:   1545  N14228    EWR  IAH      227     1400    5     15 2013-01-01 05:00:00
                    +
                    +# get last few rows
+tail(flights.df, 1)
+#>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier
+#> 1: 2013     9  30       NA            840        NA       NA           1020        NA      MQ
+#>    flight tailnum origin dest air_time distance hour minute           time_hour
+#> 1:   3531  N839MQ    LGA  RDU       NA      431    8     40 2013-09-30 08:00:00
                    +
                    +# number of rows
+nrow(flights.df)
+#> [1] 336776
                    +
                    +# number of columns
+ncol(flights.df)
+#> [1] 19

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/as.data.frame.disk.frame.html b/docs/reference/as.data.frame.disk.frame.html index 77c883d0..275aebaf 100644 --- a/docs/reference/as.data.frame.disk.frame.html +++ b/docs/reference/as.data.frame.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0
                    @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,7 +166,7 @@

                      Convert disk.frame to data.frame by collecting all chunks

                    • # S3 method for disk.frame
-as.data.frame(x, row.names, optional, ...)
                    +as.data.frame(x, row.names, optional, ...)

                    Arguments

                    @@ -188,8 +191,9 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-as.data.frame(cars.df)
                    #>    speed dist
+    
                    cars.df = as.disk.frame(cars)
+as.data.frame(cars.df)
+
                    #>    speed dist
 #> 1      4    2
 #> 2      4   10
 #> 3      7    4
@@ -241,7 +245,8 @@ 
                    Examp
 #> 49    24  120
 #> 50    25   85
                    
 # clean up
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/as.data.table.disk.frame.html b/docs/reference/as.data.table.disk.frame.html index a705e8c7..16530f9e 100644 --- a/docs/reference/as.data.table.disk.frame.html +++ b/docs/reference/as.data.table.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,7 +166,7 @@

                      Convert disk.frame to data.table by collecting all chunks

                      # S3 method for disk.frame
-as.data.table(x, keep.rownames = FALSE, ...)
                      +as.data.table(x, keep.rownames = FALSE, ...)

                      Arguments

                    • @@ -184,13 +187,15 @@

                    Arg

                    Examples

                    - 
                    library(data.table)
                    #> 
+    
                    library(data.table)
+
                    #> 
 #> Attaching package: 'data.table'
                    #> The following object is masked from 'package:purrr':
 #> 
 #>     transpose
                    #> The following objects are masked from 'package:dplyr':
 #> 
-#>     between, first, last
                    cars.df = as.disk.frame(cars)
-as.data.table(cars.df)
                    #>     speed dist
+#>     between, first, last
                    cars.df = as.disk.frame(cars)
+as.data.table(cars.df)
+
                    #>     speed dist
 #>  1:     4    2
 #>  2:     4   10
 #>  3:     7    4
@@ -243,7 +248,8 @@ 
                    Examp
 #> 50:    25   85
 #>     speed dist
                    
 # clean up
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/as.disk.frame.html b/docs/reference/as.disk.frame.html index 6fe5d859..b6f779b8 100644 --- a/docs/reference/as.disk.frame.html +++ b/docs/reference/as.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,15 +165,15 @@

                      Make a data.frame into a disk.frame

                      Make a data.frame into a disk.frame

                      - 
                      as.disk.frame(
-  df,
-  outdir = tempfile(fileext = ".df"),
-  nchunks = recommend_nchunks(df),
-  overwrite = FALSE,
-  shardby = NULL,
-  compress = 50,
-  ...
-)
                      + 
                      as.disk.frame(
+  df,
+  outdir = tempfile(fileext = ".df"),
+  nchunks = recommend_nchunks(df),
+  overwrite = FALSE,
+  shardby = NULL,
+  compress = 50,
+  ...
+)

                      Arguments

                    • @@ -208,19 +211,20 @@

                    Arg

                    Examples

                    # write to temporary location
-cars.df = as.disk.frame(cars)
+cars.df = as.disk.frame(cars) 
 
 # specify a different path in the temporary folder, you are free to choose a different folder
-cars_new_location.df = as.disk.frame(cars, outdir = file.path(tempdir(), "some_path.df"))
+cars_new_location.df = as.disk.frame(cars, outdir = file.path(tempdir(), "some_path.df"))
 
 # specify a different number of chunks
 # this writes to tempdir() by default
-cars_chunks.df = as.disk.frame(cars, nchunks = 4, overwrite = TRUE)
+cars_chunks.df = as.disk.frame(cars, nchunks = 4, overwrite = TRUE) 
 
 # clean up
-delete(cars.df)
-delete(cars_new_location.df)
-delete(cars_chunks.df)
                    +delete(cars.df) +delete(cars_new_location.df) +delete(cars_chunks.df) +
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/chunk_group_by.html b/docs/reference/chunk_group_by.html index 97d24c23..731d94e4 100644 --- a/docs/reference/chunk_group_by.html +++ b/docs/reference/chunk_group_by.html @@ -75,7 +75,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -83,7 +83,7 @@
                  • @@ -139,7 +142,7 @@
                    • - +
                      • @@ -168,13 +171,13 @@

                      Group by within each disk.frame

                      reorganizes the chunks by the shard key.

                      - 
                      chunk_summarize(.data, ...)
+    
                      chunk_summarize(.data, ...)
 
-chunk_summarise(.data, ...)
+chunk_summarise(.data, ...)
 
-chunk_group_by(.data, ...)
+chunk_group_by(.data, ...)
 
-chunk_ungroup(.data, ...)
                      
+chunk_ungroup(.data, ...)

                      Arguments

                    • @@ -208,7 +211,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/cmap.html b/docs/reference/cmap.html index 06e4a674..c2cd39fa 100644 --- a/docs/reference/cmap.html +++ b/docs/reference/cmap.html @@ -76,7 +76,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -84,7 +84,7 @@
                  • @@ -140,7 +143,7 @@
                    • - +
                      • @@ -170,91 +173,91 @@

                      Apply the same function to all chunks

                      `delayed` is an alias for lazy and is consistent with the naming in Dask and Dagger.jl

                      - 
                      cmap(.x, .f, ...)
+    
                      cmap(.x, .f, ...)
 
 # S3 method for disk.frame
-cmap(
-  .x,
-  .f,
-  ...,
-  outdir = NULL,
-  keep = NULL,
-  chunks = nchunks(.x),
-  compress = 50,
-  lazy = TRUE,
-  overwrite = FALSE,
-  vars_and_pkgs = future::getGlobalsAndPackages(.f, envir = parent.frame()),
-  .progress = TRUE
-)
-
-cmap_dfr(.x, .f, ..., .id = NULL)
+cmap(
+  .x,
+  .f,
+  ...,
+  outdir = NULL,
+  keep = NULL,
+  chunks = nchunks(.x),
+  compress = 50,
+  lazy = TRUE,
+  overwrite = FALSE,
+  vars_and_pkgs = future::getGlobalsAndPackages(.f, envir = parent.frame()),
+  .progress = TRUE
+)
+
+cmap_dfr(.x, .f, ..., .id = NULL)
 
 # S3 method for disk.frame
-cmap_dfr(.x, .f, ..., .id = NULL, use.names = fill, fill = FALSE, idcol = NULL)
+cmap_dfr(.x, .f, ..., .id = NULL, use.names = fill, fill = FALSE, idcol = NULL)
 
-cimap(.x, .f, ...)
+cimap(.x, .f, ...)
 
 # S3 method for disk.frame
-cimap(
-  .x,
-  .f,
-  outdir = NULL,
-  keep = NULL,
-  chunks = nchunks(.x),
-  compress = 50,
-  lazy = TRUE,
-  overwrite = FALSE,
-  ...
-)
-
-cimap_dfr(.x, .f, ..., .id = NULL)
+cimap(
+  .x,
+  .f,
+  outdir = NULL,
+  keep = NULL,
+  chunks = nchunks(.x),
+  compress = 50,
+  lazy = TRUE,
+  overwrite = FALSE,
+  ...
+)
+
+cimap_dfr(.x, .f, ..., .id = NULL)
 
 # S3 method for disk.frame
-cimap_dfr(
-  .x,
-  .f,
-  ...,
-  .id = NULL,
-  use.names = fill,
-  fill = FALSE,
-  idcol = NULL
-)
-
-lazy(.x, .f, ...)
+cimap_dfr(
+  .x,
+  .f,
+  ...,
+  .id = NULL,
+  use.names = fill,
+  fill = FALSE,
+  idcol = NULL
+)
+
+lazy(.x, .f, ...)
 
 # S3 method for disk.frame
-lazy(.x, .f, ...)
+lazy(.x, .f, ...)
 
-delayed(.x, .f, ...)
+delayed(.x, .f, ...)
 
-chunk_lapply(...)
+chunk_lapply(...)
 
-map(.x, .f, ...)
+map(.x, .f, ...)
 
 # S3 method for disk.frame
-map(...)
+map(...)
 
 # S3 method for default
-map(.x, .f, ...)
+map(.x, .f, ...)
 
-imap_dfr(.x, .f, ..., .id = NULL)
+imap_dfr(.x, .f, ..., .id = NULL)
 
 # S3 method for disk.frame
-imap_dfr(...)
+imap_dfr(...)
 
 # S3 method for default
-imap_dfr(.x, .f, ..., .id = NULL)
+imap_dfr(.x, .f, ..., .id = NULL)
 
-imap(.x, .f, ...)
+imap(.x, .f, ...)
 
 # S3 method for default
-imap(.x, .f, ...)
+imap(.x, .f, ...)
 
 # S3 method for disk.frame
-map_dfr(...)
+map_dfr(...)
 
 # S3 method for default
-map_dfr(.x, .f, ..., .id = NULL)
                      
+map_dfr(.x, .f, ..., .id = NULL)

                      Arguments

                    • @@ -323,15 +326,16 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # return the first row of each chunk lazily
 # 
-cars2 = cmap(cars.df, function(chunk) {
- chunk[,1]
-})
+cars2 = cmap(cars.df, function(chunk) {
+ chunk[,1]
+})
 
-collect(cars2)
                    #>     speed
+collect(cars2)
+
                    #>     speed
 #>  1:     4
 #>  2:     4
 #>  3:     7
@@ -384,9 +388,10 @@ 
                    Examp
 #> 50:    25
 #>     speed
                    
 # same as above but using purrr 
-cars2 = cmap(cars.df, ~.x[1,])
+cars2 = cmap(cars.df, ~.x[1,])
 
-collect(cars2)
                    #>    speed dist
+collect(cars2)
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:    11   17
 #> 3:    13   46
@@ -394,7 +399,8 @@ 
                    Examp
 #> 5:    19   46
 #> 6:    24   70
                    
 # return the first row of each chunk eagerly as list
-cmap(cars.df, ~.x[1,], lazy = FALSE)
                    #> [[1]]
+cmap(cars.df, ~.x[1,], lazy = FALSE)
+
                    #> [[1]]
 #>    speed dist
 #> 1:     4    2
 #> 
@@ -419,7 +425,8 @@ 
                    Examp
 #> 1:    24   70
 #> 
                    
 # return the first row of each chunk eagerly as data.table/data.frame by row-binding
-cmap_dfr(cars.df, ~.x[1,])
                    #>    speed dist
+cmap_dfr(cars.df, ~.x[1,])
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:    11   17
 #> 3:    13   46
@@ -427,13 +434,15 @@ 
                    Examp
 #> 5:    19   46
 #> 6:    24   70
                    
 # lazy and delayed are just an aliases for cmap(..., lazy = TRUE)
-collect(lazy(cars.df, ~.x[1,]))
                    #>    speed dist
+collect(lazy(cars.df, ~.x[1,]))
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:    11   17
 #> 3:    13   46
 #> 4:    16   40
 #> 5:    19   46
-#> 6:    24   70
                    collect(delayed(cars.df, ~.x[1,]))
                    #>    speed dist
+#> 6:    24   70
                    collect(delayed(cars.df, ~.x[1,]))
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:    11   17
 #> 3:    13   46
@@ -441,13 +450,14 @@ 
                    Examp
 #> 5:    19   46
 #> 6:    24   70
                    
 # clean up cars.df
-delete(cars.df)
-cars.df = as.disk.frame(cars)
+delete(cars.df)
+cars.df = as.disk.frame(cars)
 
 # .x is the chunk and .y is the ID as an integer
 
 # lazy = TRUE support is not available at the moment
-cimap(cars.df, ~.x[, id := .y], lazy = FALSE)
                    #> [[1]]
+cimap(cars.df, ~.x[, id := .y], lazy = FALSE)
+
                    #> [[1]]
 #>    speed dist id
 #> 1:     4    2  1
 #> 2:     4   10  1
@@ -515,7 +525,8 @@ 
                    Examp
 #> 4:    24  120  6
 #> 5:    25   85  6
 #> 
                    
-cimap_dfr(cars.df, ~.x[, id := .y])
                    #>     speed dist id
+cimap_dfr(cars.df, ~.x[, id := .y])
+
                    #>     speed dist id
 #>  1:     4    2  1
 #>  2:     4   10  1
 #>  3:     7    4  1
@@ -568,7 +579,8 @@ 
                    Examp
 #> 50:    25   85  6
 #>     speed dist id
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/cmap2.html b/docs/reference/cmap2.html index 795270b3..35f65ab5 100644 --- a/docs/reference/cmap2.html +++ b/docs/reference/cmap2.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                  • @@ -137,7 +140,7 @@
                    • - +
                      • @@ -164,11 +167,11 @@

                      `cmap2` a function to two disk.frames

                      gets run by .f(x.chunk, y.chunk)

                      - 
                      cmap2(.x, .y, .f, ...)
+    
                      cmap2(.x, .y, .f, ...)
 
-map2(.x, .y, .f, ...)
+map2(.x, .y, .f, ...)
 
-map_by_chunk_id(.x, .y, .f, ..., outdir)
                      
+map_by_chunk_id(.x, .y, .f, ..., outdir)

                      Arguments

                    • @@ -198,10 +201,11 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
-cars2.df = cmap2(cars.df, cars.df, ~data.table::rbindlist(list(.x, .y)))
-collect(cars2.df)
                    #>      speed dist
+cars2.df = cmap2(cars.df, cars.df, ~data.table::rbindlist(list(.x, .y)))
+collect(cars2.df)
+
                    #>      speed dist
 #>   1:     4    2
 #>   2:     4   10
 #>   3:     7    4
@@ -304,8 +308,9 @@ 
                    Examp
 #> 100:    25   85
 #>      speed dist
                    
 # clean up cars.df
-delete(cars.df)
-delete(cars2.df)
                    
+delete(cars.df)
+delete(cars2.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/collect.html b/docs/reference/collect.html index b4759f5f..01995d45 100644 --- a/docs/reference/collect.html +++ b/docs/reference/collect.html @@ -75,7 +75,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -83,7 +83,7 @@
                  • @@ -139,7 +142,7 @@
                    • - +
                      • @@ -169,12 +172,12 @@

                      Bring the disk.frame into R

                      # S3 method for disk.frame
-collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
+collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
 
-collect_list(x, simplify = FALSE, parallel = !is.null(attr(x, "lazyfn")))
+collect_list(x, simplify = FALSE, parallel = !is.null(attr(x, "lazyfn")))
 
 # S3 method for summarized_disk.frame
-collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
                      +collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))

                      Arguments

                    • @@ -209,9 +212,10 @@

                    Value

                    collect return a data.frame/data.table

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 # use collect to bring the data into RAM as a data.table/data.frame
-collect(cars.df)
                    #>     speed dist
+collect(cars.df)
+
                    #>     speed dist
 #>  1:     4    2
 #>  2:     4   10
 #>  3:     7    4
@@ -264,11 +268,12 @@ 
                    Examp
 #> 50:    25   85
 #>     speed dist
                    
 # clean up
-delete(cars.df)
-cars.df = as.disk.frame(cars)
+delete(cars.df)
+cars.df = as.disk.frame(cars)
 
 # returns the result as a list
-collect_list(cmap(cars.df, ~1))
                    #> [[1]]
+collect_list(cmap(cars.df, ~1))
+
                    #> [[1]]
 #> [1] 1
 #> 
 #> [[2]]
@@ -287,10 +292,11 @@ 
                    Examp
 #> [1] 1
 #> 
                    
 # clean up
-delete(cars.df)
-cars.df = as.disk.frame(cars)
+delete(cars.df)
+cars.df = as.disk.frame(cars)
 # use collect to bring the data into RAM as a data.table/data.frame
-collect(cars.df)
                    #>     speed dist
+collect(cars.df)
+
                    #>     speed dist
 #>  1:     4    2
 #>  2:     4   10
 #>  3:     7    4
@@ -343,7 +349,8 @@ 
                    Examp
 #> 50:    25   85
 #>     speed dist
                    
 # clean up
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/colnames.html b/docs/reference/colnames.html index 4d019358..432d0801 100644 --- a/docs/reference/colnames.html +++ b/docs/reference/colnames.html @@ -40,9 +40,10 @@ - + @@ -74,7 +75,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -82,7 +83,7 @@
                  • @@ -138,7 +142,7 @@
                    • - +
                      • @@ -161,21 +165,22 @@

                      Return the column names of the disk.frame

                      -

                      The returned column names are from the source. So if you have lazy operations then the -colnames here does not reflects the results of those operations. To obtain the correct names try -names(collect(get_chunk(df, 1)))

                      +

                      The returned column names are from the source. So if you have lazy operations +then the colnames here does not reflects the results of those +operations. Note: if you have expensive lazy function then this operation +might take some time.

                      - 
                      colnames(x, ...)
+    
                      colnames(x, ...)
 
 # S3 method for disk.frame
-names(x, ...)
+names(x, ...)
 
 # S3 method for disk.frame
-colnames(x, ...)
+colnames(x, ...)
 
 # S3 method for default
-colnames(x, ...)
                      
+colnames(x, ...)

                      Arguments

                    • @@ -206,7 +211,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/compute.disk.frame.html b/docs/reference/compute.disk.frame.html index 8204b014..9c62247e 100644 --- a/docs/reference/compute.disk.frame.html +++ b/docs/reference/compute.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,13 +166,13 @@

                      Compute without writing

                      # S3 method for disk.frame
-compute(
-  x,
-  name,
-  outdir = tempfile("tmp_df_", fileext = ".df"),
-  overwrite = TRUE,
-  ...
-)
                      +compute( + x, + name, + outdir = tempfile("tmp_df_", fileext = ".df"), + overwrite = TRUE, + ... +)

                      Arguments

                    • @@ -198,14 +201,15 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-cars.df2 = cars.df %>% cmap(~.x)
+    
                    cars.df = as.disk.frame(cars)
+cars.df2 = cars.df %>% cmap(~.x)
 # the computation is performed and the data is now stored elsewhere
-cars.df3 = compute(cars.df2)
+cars.df3 = compute(cars.df2)
 
 # clean up
-delete(cars.df)
-delete(cars.df3)
                    
+delete(cars.df)
+delete(cars.df3)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/create_chunk_mapper.html b/docs/reference/create_chunk_mapper.html index 9fc66aac..041c6bb6 100644 --- a/docs/reference/create_chunk_mapper.html +++ b/docs/reference/create_chunk_mapper.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Create function that applies to each chunk if disk.frame

                      A function to make it easier to create functions like filter

                      - 
                      create_chunk_mapper(chunk_fn, warning_msg = NULL, as.data.frame = TRUE)
                      + 
                      create_chunk_mapper(chunk_fn, warning_msg = NULL, as.data.frame = TRUE)

                      Arguments

                    • @@ -184,57 +187,61 @@

                    Arg

                    Examples

                    
-filter = create_chunk_mapper(dplyr::filter)
+filter = create_chunk_mapper(dplyr::filter)
 
 #' example: creating a function that keeps only the first and last n row
-first_and_last <- function(chunk, n, ...) {
-  nr = nrow(chunk)
-  print(nr-n+1:nr)
-  chunk[c(1:n, (nr-n+1):nr), ]
-}
+first_and_last <- function(chunk, n, ...) {
+  nr = nrow(chunk)
+  print(nr-n+1:nr)
+  chunk[c(1:n, (nr-n+1):nr), ]
+}
 
 #' create the function for use with disk.frame
-first_and_last_df = create_chunk_mapper(first_and_last)
+first_and_last_df = create_chunk_mapper(first_and_last)
 
-mtcars.df = as.disk.frame(mtcars)
+mtcars.df = as.disk.frame(mtcars)
 
 #' the operation is lazy
-lazy_mtcars.df = mtcars.df %>%
-  first_and_last_df(2)
+lazy_mtcars.df = mtcars.df %>%
+  first_and_last_df(2)
 
 #' bring into R
-collect(lazy_mtcars.df)
                    #> [1]  5  6  7  8  9 10
+collect(lazy_mtcars.df)
+
                    #> [1]  5  6  7  8  9 10
 #> [1]  5  6  7  8  9 10
 #> [1]  5  6  7  8  9 10
 #> [1]  5  6  7  8  9 10
 #> [1]  5  6  7  8  9 10
-#> [1] 1 2
                    #>        mpg cyl  disp  hp drat    wt  qsec vs am gear carb
-#> ...1  21.0   6 160.0 110 3.90 2.620 16.46  0  1    4    4
-#> ...2  21.0   6 160.0 110 3.90 2.875 17.02  0  1    4    4
-#> ...3  18.7   8 360.0 175 3.15 3.440 17.02  0  0    3    2
-#> ...4  18.1   6 225.0 105 2.76 3.460 20.22  1  0    3    1
-#> ...5  14.3   8 360.0 245 3.21 3.570 15.84  0  0    3    4
-#> ...6  24.4   4 146.7  62 3.69 3.190 20.00  1  0    4    2
-#> ...7  17.8   6 167.6 123 3.92 3.440 18.90  1  0    4    4
-#> ...8  16.4   8 275.8 180 3.07 4.070 17.40  0  0    3    3
-#> ...9  17.3   8 275.8 180 3.07 3.730 17.60  0  0    3    3
-#> ...10 15.2   8 275.8 180 3.07 3.780 18.00  0  0    3    3
-#> ...11 14.7   8 440.0 230 3.23 5.345 17.42  0  0    3    4
-#> ...12 32.4   4  78.7  66 4.08 2.200 19.47  1  1    4    1
-#> ...13 30.4   4  75.7  52 4.93 1.615 18.52  1  1    4    2
-#> ...14 33.9   4  71.1  65 4.22 1.835 19.90  1  1    4    1
-#> ...15 15.2   8 304.0 150 3.15 3.435 17.30  0  0    3    2
-#> ...16 13.3   8 350.0 245 3.73 3.840 15.41  0  0    3    4
-#> ...17 19.2   8 400.0 175 3.08 3.845 17.05  0  0    3    2
-#> ...18 27.3   4  79.0  66 4.08 1.935 18.90  1  1    4    1
-#> ...19 15.8   8 351.0 264 4.22 3.170 14.50  0  1    5    4
-#> ...20 19.7   6 145.0 175 3.62 2.770 15.50  0  1    5    6
-#> 1     15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
-#> 2     21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
-#> 1.1   15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
-#> 2.1   21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
                    
+#> [1] 1 2
                    #>      mpg cyl  disp  hp drat    wt  qsec vs am gear carb
+#>  1: 21.0   6 160.0 110 3.90 2.620 16.46  0  1    4    4
+#>  2: 21.0   6 160.0 110 3.90 2.875 17.02  0  1    4    4
+#>  3: 18.7   8 360.0 175 3.15 3.440 17.02  0  0    3    2
+#>  4: 18.1   6 225.0 105 2.76 3.460 20.22  1  0    3    1
+#>  5: 14.3   8 360.0 245 3.21 3.570 15.84  0  0    3    4
+#>  6: 24.4   4 146.7  62 3.69 3.190 20.00  1  0    4    2
+#>  7: 17.8   6 167.6 123 3.92 3.440 18.90  1  0    4    4
+#>  8: 16.4   8 275.8 180 3.07 4.070 17.40  0  0    3    3
+#>  9: 17.3   8 275.8 180 3.07 3.730 17.60  0  0    3    3
+#> 10: 15.2   8 275.8 180 3.07 3.780 18.00  0  0    3    3
+#> 11: 14.7   8 440.0 230 3.23 5.345 17.42  0  0    3    4
+#> 12: 32.4   4  78.7  66 4.08 2.200 19.47  1  1    4    1
+#> 13: 30.4   4  75.7  52 4.93 1.615 18.52  1  1    4    2
+#> 14: 33.9   4  71.1  65 4.22 1.835 19.90  1  1    4    1
+#> 15: 15.2   8 304.0 150 3.15 3.435 17.30  0  0    3    2
+#> 16: 13.3   8 350.0 245 3.73 3.840 15.41  0  0    3    4
+#> 17: 19.2   8 400.0 175 3.08 3.845 17.05  0  0    3    2
+#> 18: 27.3   4  79.0  66 4.08 1.935 18.90  1  1    4    1
+#> 19: 15.8   8 351.0 264 4.22 3.170 14.50  0  1    5    4
+#> 20: 19.7   6 145.0 175 3.62 2.770 15.50  0  1    5    6
+#> 21: 15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
+#> 22: 21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
+#> 23: 15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
+#> 24: 21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
+#>      mpg cyl  disp  hp drat    wt  qsec vs am gear carb
                    
 #' clean up
-delete(mtcars.df)
                    +delete(mtcars.df) + +
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/csv_to_disk.frame.html b/docs/reference/csv_to_disk.frame.html index 3f80f571..687aa9d8 100644 --- a/docs/reference/csv_to_disk.frame.html +++ b/docs/reference/csv_to_disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,21 +165,21 @@

                      Convert CSV file(s) to disk.frame format

                      Convert CSV file(s) to disk.frame format

                      - 
                      csv_to_disk.frame(
-  infile,
-  outdir = tempfile(fileext = ".df"),
-  inmapfn = base::I,
-  nchunks = recommend_nchunks(sum(file.size(infile))),
-  in_chunk_size = NULL,
-  shardby = NULL,
-  compress = 50,
-  overwrite = TRUE,
-  header = TRUE,
-  .progress = TRUE,
-  backend = c("data.table", "readr", "LaF"),
-  chunk_reader = c("bigreadr", "data.table", "readr", "readLines"),
-  ...
-)
                      + 
                      csv_to_disk.frame(
+  infile,
+  outdir = tempfile(fileext = ".df"),
+  inmapfn = base::I,
+  nchunks = recommend_nchunks(sum(file.size(infile))),
+  in_chunk_size = NULL,
+  shardby = NULL,
+  compress = 50,
+  overwrite = TRUE,
+  header = TRUE,
+  .progress = TRUE,
+  backend = c("data.table", "readr", "LaF"),
+  chunk_reader = c("bigreadr", "data.table", "readr", "readLines"),
+  ...
+)

                      Arguments

                    • @@ -264,14 +267,15 @@

                    See a zip_to_disk.frame()

                    Examples

                    - 
                    tmpfile = tempfile()
-write.csv(cars, tmpfile)
-tmpdf = tempfile(fileext = ".df")
-df = csv_to_disk.frame(tmpfile, outdir = tmpdf, overwrite = TRUE)
+    
                    tmpfile = tempfile()
+write.csv(cars, tmpfile)
+tmpdf = tempfile(fileext = ".df")
+df = csv_to_disk.frame(tmpfile, outdir = tmpdf, overwrite = TRUE)
 
 # clean up
-fs::file_delete(tmpfile)
-delete(df)
                    
+fs::file_delete(tmpfile)
+delete(df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/delete.html b/docs/reference/delete.html index f44d0caf..d890c7ba 100644 --- a/docs/reference/delete.html +++ b/docs/reference/delete.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Delete a disk.frame

                      Delete a disk.frame

                      - 
                      delete(df)
                      + 
                      delete(df)

                      Arguments

                    • @@ -175,8 +178,9 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-delete(cars.df)
                    + 
                    cars.df = as.disk.frame(cars)
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/df_ram_size.html b/docs/reference/df_ram_size.html index 6a7c7b25..c3ac569a 100644 --- a/docs/reference/df_ram_size.html +++ b/docs/reference/df_ram_size.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Get the size of RAM in gigabytes

                      Get the size of RAM in gigabytes

                      - 
                      df_ram_size()
                      + 
                      df_ram_size()

                      Value

                      @@ -171,7 +174,8 @@

                      Value

                      Examples

                      # returns the RAM size in gigabyte (GB)
-df_ram_size()
                      #> [1] 64
                      +df_ram_size() +
                      #> [1] 64
                      -

                      Site built with pkgdown 1.5.1.

                      +

                      Site built with pkgdown 1.6.1.

                      diff --git a/docs/reference/dfglm.html b/docs/reference/dfglm.html index 75c7571c..ee6fc36b 100644 --- a/docs/reference/dfglm.html +++ b/docs/reference/dfglm.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                    • @@ -137,7 +140,7 @@
                      • - +
                        • @@ -164,7 +167,7 @@

                        Fit generalized linear models (glm) with disk.frame

                        those return by those functions. This is a convenience wrapper

                        - 
                        dfglm(formula, data, ..., glm_backend = c("biglm", "speedglm", "biglmm"))
                        + 
                        dfglm(formula, data, ..., glm_backend = c("biglm", "speedglm", "biglmm"))

                        Arguments

                    • @@ -224,25 +227,28 @@

                    See a make_glm_streaming_fn()

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-m = dfglm(dist ~ speed, data = cars.df)
                    #> Loading required namespace: biglm
                    
+    
                    cars.df = as.disk.frame(cars)
+m = dfglm(dist ~ speed, data = cars.df)
+
                    #> Loading required namespace: biglm
                    
 # can use normal R functions
 # Only works in version > R 3.6
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
-if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
-  summary(m)
-  predict(m, get_chunk(cars.df, 1))
-  predict(m, collect(cars.df))
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
+  summary(m)
+  predict(m, get_chunk(cars.df, 1))
+  predict(m, collect(cars.df))
   # can use broom to tidy up the returned info
-  broom::tidy(m)
-}
                    #> # A tibble: 2 x 4
+  broom::tidy(m)
+}
+
                    #> # A tibble: 2 x 4
 #>   term        estimate std.error  p.value
 #>   <chr>          <dbl>     <dbl>    <dbl>
 #> 1 (Intercept)   -17.6      6.76  9.29e- 3
 #> 2 speed           3.93     0.416 2.96e-21
                    
 # clean up
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/disk.frame.html b/docs/reference/disk.frame.html index 5dd9de15..574c9440 100644 --- a/docs/reference/disk.frame.html +++ b/docs/reference/disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Create a disk.frame from a folder

                      Create a disk.frame from a folder

                      - 
                      disk.frame(path, backend = "fst")
                      + 
                      disk.frame(path, backend = "fst")

                      Arguments

                    • @@ -179,21 +182,25 @@

                    Arg

                    Examples

                    - 
                    path = file.path(tempdir(),"cars")
-as.disk.frame(cars, outdir=path, overwrite = TRUE, nchunks = 2)
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK/cars"
+    
                    path = file.path(tempdir(),"cars")
+as.disk.frame(cars, outdir=path, overwrite = TRUE, nchunks = 2)
+
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP/cars"
 #> nchunks: 2
 #> nrow (at source): 50
 #> ncol (at source): 2
 #> nrow (post operations): ???
-#> ncol (post operations): ???
                    df = disk.frame(path)
-head(df)
                    #>    speed dist
+#> ncol (post operations): ???
                    df = disk.frame(path)
+head(df)
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:     4   10
 #> 3:     7    4
 #> 4:     7   22
 #> 5:     8   16
-#> 6:     9   10
                    nchunks(df)
                    #> [1] 2
                    # clean up
-delete(df)
                    
+#> 6:     9   10
                    nchunks(df)
+
                    #> [1] 2
                    # clean up
+delete(df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/dplyr_verbs.html b/docs/reference/dplyr_verbs.html index 00f29a9e..9f86f10e 100644 --- a/docs/reference/dplyr_verbs.html +++ b/docs/reference/dplyr_verbs.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                  • @@ -137,7 +140,7 @@
                    • - +
                      • @@ -165,43 +168,37 @@

                      The dplyr verbs implemented for disk.frame

                      # S3 method for disk.frame
-select(.data, ...)
+select(.data, ...)
 
 # S3 method for disk.frame
-rename(.data, ...)
+rename(.data, ...)
 
 # S3 method for disk.frame
-filter(.data, ...)
+filter(.data, ...)
 
 # S3 method for disk.frame
-mutate(.data, ...)
+mutate(.data, ...)
 
 # S3 method for disk.frame
-transmute(.data, ...)
+transmute(.data, ...)
 
 # S3 method for disk.frame
-arrange(.data, ...)
-
-chunk_arrange(.data, ...)
-
-tally.disk.frame(.data, ...)
-
-count.disk.frame(.data, ...)
+arrange(.data, ...)
 
-add_count.disk.frame(.data, ...)
+chunk_arrange(.data, ...)
 
-add_tally.disk.frame(.data, ...)
+add_tally.disk.frame(.data, ...)
 
 # S3 method for disk.frame
-do(.data, ...)
+do(.data, ...)
 
 # S3 method for disk.frame
-distinct(...)
+distinct(...)
 
-chunk_distinct(.data, ...)
+chunk_distinct(.data, ...)
 
 # S3 method for disk.frame
-glimpse(.data, ...)
                      +glimpse(.data, ...)

                      Arguments

                    • @@ -218,20 +215,21 @@

                    Arg

                    Examples

                    - 
                    library(dplyr)
-cars.df = as.disk.frame(cars)
-mult = 2
+    
                    library(dplyr)
+cars.df = as.disk.frame(cars)
+mult = 2
 
 # use all any of the supported dplyr
-cars2 = cars.df %>%
-  select(speed) %>%
-  mutate(speed2 = speed * mult) %>%
-  filter(speed < 50) %>%
-  rename(speed1 = speed) %>%
-  collect
+cars2 = cars.df %>% 
+  select(speed) %>% 
+  mutate(speed2 = speed * mult) %>% 
+  filter(speed < 50) %>% 
+  rename(speed1 = speed) %>% 
+  collect
 
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/evalparseglue.html b/docs/reference/evalparseglue.html index 60206d23..ef02aa33 100644 --- a/docs/reference/evalparseglue.html +++ b/docs/reference/evalparseglue.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Helper function to evalparse some `glue::glue` string

                      Helper function to evalparse some `glue::glue` string

                      - 
                      evalparseglue(code, env = parent.frame())
                      + 
                      evalparseglue(code, env = parent.frame())

                      Arguments

                    • @@ -193,7 +196,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/foverlaps.disk.frame.html b/docs/reference/foverlaps.disk.frame.html index 11ca11f6..df3534db 100644 --- a/docs/reference/foverlaps.disk.frame.html +++ b/docs/reference/foverlaps.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,18 +165,18 @@

                      Apply data.table's foverlaps to the disk.frame

                      EXPERIMENTAL

                      - 
                      foverlaps.disk.frame(
-  df1,
-  df2,
-  by.x = if (identical(shardkey(df1)$shardkey, "")) shardkey(df1)$shardkey else
-    shardkey(df2)$shardkey,
-  by.y = shardkey(df2)$shardkey,
-  ...,
-  outdir = tempfile("df_foverlaps_tmp", fileext = ".df"),
-  merge_by_chunk_id = FALSE,
-  compress = 50,
-  overwrite = TRUE
-)
                      + 
                      foverlaps.disk.frame(
+  df1,
+  df2,
+  by.x = if (identical(shardkey(df1)$shardkey, "")) shardkey(df1)$shardkey else
+    shardkey(df2)$shardkey,
+  by.y = shardkey(df2)$shardkey,
+  ...,
+  outdir = tempfile("df_foverlaps_tmp", fileext = ".df"),
+  merge_by_chunk_id = FALSE,
+  compress = 50,
+  overwrite = TRUE
+)

                      Arguments

                    • @@ -218,20 +221,20 @@

                    Arg

                    Examples

                    - 
                    library(data.table)
+    
                    library(data.table)
 
 ## simple example:
-x = as.disk.frame(data.table(start=c(5,31,22,16), end=c(8,50,25,18), val2 = 7:10))
-y = as.disk.frame(data.table(start=c(10, 20, 30), end=c(15, 35, 45), val1 = 1:3))
-byxy = c("val1", "start", "end")
-xy.df = foverlaps.disk.frame(
-  x, y, by.x = byxy, by.y = byxy,
-  merge_by_chunk_id = TRUE, overwrite = TRUE)
-
+x = as.disk.frame(data.table(start=c(5,31,22,16), end=c(8,50,25,18), val2 = 7:10))
+y = as.disk.frame(data.table(start=c(10, 20, 30), end=c(15, 35, 45), val1 = 1:3))
+byxy = c("start", "end")
+xy.df = foverlaps.disk.frame(
+   x, y, by.x = byxy, by.y = byxy,
+  merge_by_chunk_id = TRUE, overwrite = TRUE)
 # clean up
-delete(x)
-delete(y)
-delete(xy.df)
                    
+delete(x)
+delete(y)
+delete(xy.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/gen_datatable_synthetic.html b/docs/reference/gen_datatable_synthetic.html index c7eb7d31..9a068fca 100644 --- a/docs/reference/gen_datatable_synthetic.html +++ b/docs/reference/gen_datatable_synthetic.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Generate synthetic dataset for testing

                      Generate synthetic dataset for testing

                      - 
                      gen_datatable_synthetic(N = 2e+08, K = 100)
                      + 
                      gen_datatable_synthetic(N = 2e+08, K = 100)

                      Arguments

                    • @@ -193,7 +196,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/get_chunk.html b/docs/reference/get_chunk.html index 5cf436cb..517e982e 100644 --- a/docs/reference/get_chunk.html +++ b/docs/reference/get_chunk.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,10 +165,10 @@

                      Obtain one chunk by chunk id

                      Obtain one chunk by chunk id

                      - 
                      get_chunk(...)
+    
                      get_chunk(...)
 
 # S3 method for disk.frame
-get_chunk(df, n, keep = NULL, full.names = FALSE, ...)
                      
+get_chunk(df, n, keep = NULL, full.names = FALSE, ...)

                      Arguments

                    • @@ -194,8 +197,9 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars, nchunks = 2)
-get_chunk(cars.df, 1)
                    #>     speed dist
+    
                    cars.df = as.disk.frame(cars, nchunks = 2)
+get_chunk(cars.df, 1)
+
                    #>     speed dist
 #>  1:     4    2
 #>  2:     4   10
 #>  3:     7    4
@@ -221,7 +225,8 @@ 
                    Examp
 #> 23:    14   80
 #> 24:    15   20
 #> 25:    15   26
-#>     speed dist
                    get_chunk(cars.df, 2)
                    #>     speed dist
+#>     speed dist
                    get_chunk(cars.df, 2)
+
                    #>     speed dist
 #>  1:    15   54
 #>  2:    16   32
 #>  3:    16   40
@@ -247,7 +252,8 @@ 
                    Examp
 #> 23:    24   93
 #> 24:    24  120
 #> 25:    25   85
-#>     speed dist
                    get_chunk(cars.df, 1, keep = "speed")
                    #>     speed
+#>     speed dist
                    get_chunk(cars.df, 1, keep = "speed")
+
                    #>     speed
 #>  1:     4
 #>  2:     4
 #>  3:     7
@@ -275,7 +281,8 @@ 
                    Examp
 #> 25:    15
 #>     speed
                    
 # if full.names = TRUE then the full path to the chunk need to be provided
-get_chunk(cars.df, file.path(attr(cars.df, "path"), "1.fst"), full.names = TRUE)
                    #>     speed dist
+get_chunk(cars.df, file.path(attr(cars.df, "path"), "1.fst"), full.names = TRUE)
+
                    #>     speed dist
 #>  1:     4    2
 #>  2:     4   10
 #>  3:     7    4
@@ -303,7 +310,8 @@ 
                    Examp
 #> 25:    15   26
 #>     speed dist
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/get_chunk_ids.html b/docs/reference/get_chunk_ids.html index 968aaa4b..b6b80fac 100644 --- a/docs/reference/get_chunk_ids.html +++ b/docs/reference/get_chunk_ids.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Get the chunk IDs and files names

                      Get the chunk IDs and files names

                      - 
                      get_chunk_ids(df, ..., full.names = FALSE, strip_extension = TRUE)
                      + 
                      get_chunk_ids(df, ..., full.names = FALSE, strip_extension = TRUE)

                      Arguments

                    • @@ -187,21 +190,25 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # return the integer-string chunk IDs
-get_chunk_ids(cars.df)
                    #> [1] "1" "2" "3" "4" "5" "6"
                    
+get_chunk_ids(cars.df)
+
                    #> [1] "1" "2" "3" "4" "5" "6"
                    
 # return the file name chunk IDs
-get_chunk_ids(cars.df, full.names = TRUE)
                    #> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpInritK\\file187ce9d748c.df/1.fst"
-#> [2] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpInritK\\file187ce9d748c.df/2.fst"
-#> [3] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpInritK\\file187ce9d748c.df/3.fst"
-#> [4] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpInritK\\file187ce9d748c.df/4.fst"
-#> [5] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpInritK\\file187ce9d748c.df/5.fst"
-#> [6] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpInritK\\file187ce9d748c.df/6.fst"
                    
+get_chunk_ids(cars.df, full.names = TRUE)
+
                    #> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/1.fst"
+#> [2] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/2.fst"
+#> [3] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/3.fst"
+#> [4] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/4.fst"
+#> [5] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/5.fst"
+#> [6] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/6.fst"
                    
 # return the file name chunk IDs with file extension
-get_chunk_ids(cars.df, strip_extension = FALSE)
                    #> [1] "1.fst" "2.fst" "3.fst" "4.fst" "5.fst" "6.fst"
                    
+get_chunk_ids(cars.df, strip_extension = FALSE)
+
                    #> [1] "1.fst" "2.fst" "3.fst" "4.fst" "5.fst" "6.fst"
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/group_by.html b/docs/reference/group_by.html index c2f6087d..f4e80c6e 100644 --- a/docs/reference/group_by.html +++ b/docs/reference/group_by.html @@ -75,7 +75,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -83,7 +83,7 @@
                  • @@ -139,7 +142,7 @@
                    • - +
                      • @@ -169,19 +172,19 @@

                      A function to parse the summarize function

                      # S3 method for grouped_disk.frame
-summarise(.data, ...)
+summarise(.data, ...)
 
 # S3 method for grouped_disk.frame
-summarize(.data, ...)
+summarize(.data, ...)
 
 # S3 method for disk.frame
-group_by(.data, ..., add = FALSE, .drop = dplyr::group_by_drop_default(.data))
+group_by(.data, ..., add = FALSE, .drop = dplyr::group_by_drop_default(.data))
 
 # S3 method for disk.frame
-summarize(.data, ...)
+summarize(.data, ...)
 
 # S3 method for disk.frame
-summarise(.data, ...)
                      +summarise(.data, ...)

                      Arguments

                    • @@ -223,7 +226,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/groups.disk.frame.html b/docs/reference/groups.disk.frame.html index dbe0e2dd..9c8d9c05 100644 --- a/docs/reference/groups.disk.frame.html +++ b/docs/reference/groups.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,7 +166,7 @@

                      The shard keys of the disk.frame

                      # S3 method for disk.frame
-groups(x)
                      +groups(x)

                      Arguments

                    • @@ -193,7 +196,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/hard_arrange.html b/docs/reference/hard_arrange.html index cb5c473d..d31b48ee 100644 --- a/docs/reference/hard_arrange.html +++ b/docs/reference/hard_arrange.html @@ -74,7 +74,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -82,7 +82,7 @@
                  • @@ -138,7 +141,7 @@
                    • - +
                      • @@ -166,19 +169,19 @@

                      Perform a hard arrange

                      row that share the same `by` value will end up in the same chunk.

                      - 
                      hard_arrange(df, ..., add = FALSE, .drop = FALSE)
+    
                      hard_arrange(df, ..., add = FALSE, .drop = FALSE)
 
 # S3 method for data.frame
-hard_arrange(df, ...)
+hard_arrange(df, ...)
 
 # S3 method for disk.frame
-hard_arrange(
-  df,
-  ...,
-  outdir = tempfile("tmp_disk_frame_hard_arrange"),
-  nchunks = disk.frame::nchunks(df),
-  overwrite = TRUE
-)
                      
+hard_arrange(
+  df,
+  ...,
+  outdir = tempfile("tmp_disk_frame_hard_arrange"),
+  nchunks = disk.frame::nchunks(df),
+  overwrite = TRUE
+)

                      Arguments

                    • @@ -215,11 +218,13 @@

                    Arg

                    Examples

                    - 
                    iris.df = as.disk.frame(iris, nchunks = 2)
+    
                    iris.df = as.disk.frame(iris, nchunks = 2)
 
 # arrange iris.df by specifies and ensure rows with the same specifies are in the same chunk
-iris_hard.df = hard_arrange(iris.df, Species)
                    #> Appending disk.frames: 
                    
-get_chunk(iris_hard.df, 1)
                    #> # A tibble: 50 x 5
+iris_hard.df = hard_arrange(iris.df, Species)
+
                    #> Appending disk.frames: 
                    
+get_chunk(iris_hard.df, 1)
+
                    #> # A tibble: 50 x 5
 #> # Groups:   Species [1]
 #>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species  
 #>           <dbl>       <dbl>        <dbl>       <dbl> <fct>    
@@ -233,7 +238,8 @@ 
                    Examp
 #>  8          7.3         2.9          6.3         1.8 virginica
 #>  9          6.7         2.5          5.8         1.8 virginica
 #> 10          7.2         3.6          6.1         2.5 virginica
-#> # ... with 40 more rows
                    get_chunk(iris_hard.df, 2)
                    #> # A tibble: 50 x 5
+#> # ... with 40 more rows
                    get_chunk(iris_hard.df, 2)
+
                    #> # A tibble: 50 x 5
 #> # Groups:   Species [1]
 #>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species  
 #>           <dbl>       <dbl>        <dbl>       <dbl> <fct>    
@@ -249,8 +255,9 @@ 
                    Examp
 #> 10          7.2         3.6          6.1         2.5 virginica
 #> # ... with 40 more rows
                    
 # clean up cars.df
-delete(iris.df)
-delete(iris_hard.df)
                    
+delete(iris.df)
+delete(iris_hard.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/hard_group_by.html b/docs/reference/hard_group_by.html index 1b6f636c..7df1019d 100644 --- a/docs/reference/hard_group_by.html +++ b/docs/reference/hard_group_by.html @@ -74,7 +74,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -82,7 +82,7 @@
                  • @@ -138,7 +141,7 @@
                    • - +
                      • @@ -166,23 +169,23 @@

                      Perform a hard group

                      row that share the same `by` value will end up in the same chunk.

                      - 
                      hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
+    
                      hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
 
 # S3 method for data.frame
-hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
+hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
 
 # S3 method for disk.frame
-hard_group_by(
-  df,
-  ...,
-  outdir = tempfile("tmp_disk_frame_hard_group_by"),
-  nchunks = disk.frame::nchunks(df),
-  overwrite = TRUE,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL,
-  sort_split_sample_size = 100
-)
                      
+hard_group_by(
+  df,
+  ...,
+  outdir = tempfile("tmp_disk_frame_hard_group_by"),
+  nchunks = disk.frame::nchunks(df),
+  overwrite = TRUE,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL,
+  sort_split_sample_size = 100
+)

                      Arguments

                    • @@ -235,11 +238,13 @@

                    Arg

                    Examples

                    - 
                    iris.df = as.disk.frame(iris, nchunks = 2)
+    
                    iris.df = as.disk.frame(iris, nchunks = 2)
 
 # group_by iris.df by specifies and ensure rows with the same specifies are in the same chunk
-iris_hard.df = hard_group_by(iris.df, Species)
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
-get_chunk(iris_hard.df, 1)
                    #> # A tibble: 150 x 5
+iris_hard.df = hard_group_by(iris.df, Species)
+
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
+get_chunk(iris_hard.df, 1)
+
                    #> # A tibble: 150 x 5
 #> # Groups:   Species [3]
 #>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
 #>           <dbl>       <dbl>        <dbl>       <dbl> <fct>  
@@ -253,10 +258,12 @@ 
                    Examp
 #>  8          5           3.4          1.5         0.2 setosa 
 #>  9          4.4         2.9          1.4         0.2 setosa 
 #> 10          4.9         3.1          1.5         0.1 setosa 
-#> # ... with 140 more rows
                    get_chunk(iris_hard.df, 2)
                    #> Warning: The chunk NA does not exist; returning an empty data.table
                    #> Null data.table (0 rows and 0 cols)
                    
+#> # ... with 140 more rows
                    get_chunk(iris_hard.df, 2)
+
                    #> Warning: The chunk NA does not exist; returning an empty data.table
                    #> Null data.table (0 rows and 0 cols)
                    
 # clean up cars.df
-delete(iris.df)
-delete(iris_hard.df)
                    
+delete(iris.df)
+delete(iris_hard.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/head_tail.html b/docs/reference/head_tail.html index da4adf74..74fe55c0 100644 --- a/docs/reference/head_tail.html +++ b/docs/reference/head_tail.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,10 +166,10 @@

                      Head and tail of the disk.frame

                      # S3 method for disk.frame
-head(x, n = 6L, ...)
+head(x, n = 6L, ...)
 
 # S3 method for disk.frame
-tail(x, n = 6L, ...)
                      +tail(x, n = 6L, ...)

                      Arguments

                    • @@ -187,21 +190,24 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-head(cars.df)
                    #>    speed dist
+    
                    cars.df = as.disk.frame(cars)
+head(cars.df)
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:     4   10
 #> 3:     7    4
 #> 4:     7   22
 #> 5:     8   16
-#> 6:     9   10
                    tail(cars.df)
                    #>    speed dist
+#> 6:     9   10
                    tail(cars.df)
+
                    #>    speed dist
 #> 1:    24   70
 #> 2:    24   92
 #> 3:    24   93
 #> 4:    24  120
 #> 5:    25   85
                    
 # clean up 
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/index.html b/docs/reference/index.html index 7a9bcfdc..4f823595 100644 --- a/docs/reference/index.html +++ b/docs/reference/index.html @@ -71,7 +71,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -79,7 +79,7 @@
                  • @@ -135,7 +138,7 @@
                    • - - @@ -280,7 +277,7 @@

                    select(<disk.frame>) rename(<disk.frame>) filter(<disk.frame>) mutate(<disk.frame>) transmute(<disk.frame>) arrange(<disk.frame>) chunk_arrange() tally.disk.frame() count.disk.frame() add_count.disk.frame() add_tally.disk.frame() do(<disk.frame>) distinct(<disk.frame>) chunk_distinct() glimpse(<disk.frame>)

                    +

                    select(<disk.frame>) rename(<disk.frame>) filter(<disk.frame>) mutate(<disk.frame>) transmute(<disk.frame>) arrange(<disk.frame>) chunk_arrange() add_tally.disk.frame() do(<disk.frame>) distinct(<disk.frame>) chunk_distinct() glimpse(<disk.frame>)

                    @@ -519,7 +516,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/is_disk.frame.html b/docs/reference/is_disk.frame.html index 3b294687..b64b33ea 100644 --- a/docs/reference/is_disk.frame.html +++ b/docs/reference/is_disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Checks if a folder is a disk.frame

                      Checks if a folder is a disk.frame

                      - 
                      is_disk.frame(df)
                      + 
                      is_disk.frame(df)

                      Arguments

                    • Kept for backwards-compatibility to be removed in 0.3

                    csv_to_disk.frame()

                    The dplyr verbs implemented for disk.frame
                    @@ -175,11 +178,14 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
-is_disk.frame(cars) # FALSE
                    #> [1] FALSE
                    is_disk.frame(cars.df) # TRUE
                    #> [1] TRUE
                    
+is_disk.frame(cars) # FALSE
+
                    #> [1] FALSE
                    is_disk.frame(cars.df) # TRUE
+
                    #> [1] TRUE
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/join.html b/docs/reference/join.html index 3caac2ed..3abfbc81 100644 --- a/docs/reference/join.html +++ b/docs/reference/join.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,69 +166,69 @@

                      Performs join/merge for disk.frames

                      # S3 method for disk.frame
-anti_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_anti_join"),
-  merge_by_chunk_id = FALSE,
-  overwrite = TRUE,
-  .progress = FALSE
-)
+anti_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_anti_join"),
+  merge_by_chunk_id = FALSE,
+  overwrite = TRUE,
+  .progress = FALSE
+)
 
 # S3 method for disk.frame
-full_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_full_join"),
-  overwrite = TRUE,
-  merge_by_chunk_id,
-  .progress = FALSE
-)
+full_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_full_join"),
+  overwrite = TRUE,
+  merge_by_chunk_id,
+  .progress = FALSE
+)
 
 # S3 method for disk.frame
-inner_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_inner_join"),
-  merge_by_chunk_id = NULL,
-  overwrite = TRUE,
-  .progress = FALSE
-)
+inner_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_inner_join"),
+  merge_by_chunk_id = NULL,
+  overwrite = TRUE,
+  .progress = FALSE
+)
 
 # S3 method for disk.frame
-left_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_left_join"),
-  merge_by_chunk_id = FALSE,
-  overwrite = TRUE,
-  .progress = FALSE
-)
+left_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_left_join"),
+  merge_by_chunk_id = FALSE,
+  overwrite = TRUE,
+  .progress = FALSE
+)
 
 # S3 method for disk.frame
-semi_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_semi_join"),
-  merge_by_chunk_id = FALSE,
-  overwrite = TRUE,
-  .progress = FALSE
-)
                      +semi_join( + x, + y, + by = NULL, + copy = FALSE, + ..., + outdir = tempfile("tmp_disk_frame_semi_join"), + merge_by_chunk_id = FALSE, + overwrite = TRUE, + .progress = FALSE +)

                      Arguments

                    • @@ -273,43 +276,49 @@

                    Value

                    disk.frame or data.frame/data.table

                    Examples

                    - 
                    df.df = as.disk.frame(data.frame(x = 1:3, y = 4:6), overwrite = TRUE)
-df2.df = as.disk.frame(data.frame(x = 1:2, z = 10:11), overwrite = TRUE)
+    
                    df.df = as.disk.frame(data.frame(x = 1:3, y = 4:6), overwrite = TRUE)
+df2.df = as.disk.frame(data.frame(x = 1:2, z = 10:11), overwrite = TRUE)
 
-anti_joined.df = anti_join(df.df, df2.df)
                    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
-anti_joined.df %>% collect
                    #>    x y
+anti_joined.df = anti_join(df.df, df2.df) 
+
                    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
+anti_joined.df %>% collect
+
                    #>    x y
 #> 1: 3 6
                    
-anti_joined.data.frame = anti_join(df.df, data.frame(x = 1:2, z = 10:11))
                    #> Joining, by = "x"
                    #> Joining, by = "x"
                    #> Joining, by = "x"
                    
+anti_joined.data.frame = anti_join(df.df, data.frame(x = 1:2, z = 10:11))
+
                    #> Joining, by = "x"
                    #> Joining, by = "x"
                    #> Joining, by = "x"
                    
 # clean up
-delete(df.df)
-delete(df2.df)
-delete(anti_joined.df)
-cars.df = as.disk.frame(cars)
+delete(df.df)
+delete(df2.df)
+delete(anti_joined.df)
+cars.df = as.disk.frame(cars)
 
-join.df = full_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
+join.df = full_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
 
 # clean up cars.df
-delete(cars.df)
-delete(join.df)
-cars.df = as.disk.frame(cars)
+delete(cars.df)
+delete(join.df)
+cars.df = as.disk.frame(cars)
 
-join.df = inner_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
+join.df = inner_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
 
 # clean up cars.df
-delete(cars.df)
-delete(join.df)
-cars.df = as.disk.frame(cars)
+delete(cars.df)
+delete(join.df)
+cars.df = as.disk.frame(cars)
 
-join.df = left_join(cars.df, cars.df)
                    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
+join.df = left_join(cars.df, cars.df)
+
                    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
 # clean up cars.df
-delete(cars.df)
-delete(join.df)
-cars.df = as.disk.frame(cars)
+delete(cars.df)
+delete(join.df)
+cars.df = as.disk.frame(cars)
 
-join.df = semi_join(cars.df, cars.df)
                    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
+join.df = semi_join(cars.df, cars.df)
+
                    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
 # clean up cars.df
-delete(cars.df)
-delete(join.df)
                    
+delete(cars.df)
+delete(join.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/make_glm_streaming_fn.html b/docs/reference/make_glm_streaming_fn.html index e6df5579..1f76192a 100644 --- a/docs/reference/make_glm_streaming_fn.html +++ b/docs/reference/make_glm_streaming_fn.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      A streaming function for speedglm

                      Define a function that can be used to feed data into speedglm and biglm

                      - 
                      make_glm_streaming_fn(data, verbose = FALSE)
                      + 
                      make_glm_streaming_fn(data, verbose = FALSE)

                      Arguments

                    • @@ -186,19 +189,20 @@

                    See a dfglm()

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-streamacq = make_glm_streaming_fn(cars.df, verbose = FALSE)
-
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
-if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
-  m = biglm::bigglm(dist ~ speed, data = streamacq)
-  summary(m)
-  predict(m, get_chunk(cars.df, 1))
-  predict(m, collect(cars.df, 1))
-} else {
-  m = speedglm::shglm(dist ~ speed, data = streamacq)
-}
                    #>         [,1]
+    
                    cars.df = as.disk.frame(cars)
+streamacq = make_glm_streaming_fn(cars.df, verbose = FALSE)
+
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
+  m = biglm::bigglm(dist ~ speed, data = streamacq)
+  summary(m)
+  predict(m, get_chunk(cars.df, 1))
+  predict(m, collect(cars.df, 1))
+} else {
+  m = speedglm::shglm(dist ~ speed, data = streamacq)
+}
+
                    #>         [,1]
 #> 1  -1.849460
 #> 2  -1.849460
 #> 3   9.947766
@@ -264,7 +268,7 @@ 
                    Contents
                    
 
                    
 
 
                    
-  
                    Site built with pkgdown 1.5.1.
                    
+  
                    Site built with pkgdown 1.6.1.
                    
 
                    
 
       
diff --git a/docs/reference/merge.disk.frame.html b/docs/reference/merge.disk.frame.html
index 7e9d9792..b2d35726 100644
--- a/docs/reference/merge.disk.frame.html
+++ b/docs/reference/merge.disk.frame.html
@@ -72,7 +72,7 @@
       
       
         disk.frame
-        0.3.6
+        0.5.0
       
     
                    
 
@@ -80,7 +80,7 @@
       
 
 
                  • 
@@ -136,7 +139,7 @@
       
                      
         
                    • 
   
-    
+    
      
   
 
                      • 
@@ -163,15 +166,15 @@ 
                      Merge function for disk.frames
                      
     
 
     
                      # S3 method for disk.frame
-merge(
-  x,
-  y,
-  by,
-  outdir = tempfile(fileext = ".df"),
-  ...,
-  merge_by_chunk_id = FALSE,
-  overwrite = FALSE
-)
                      
+merge(
+  x,
+  y,
+  by,
+  outdir = tempfile(fileext = ".df"),
+  ...,
+  merge_by_chunk_id = FALSE,
+  overwrite = FALSE
+)

                    • Arguments

                    @@ -208,14 +211,15 @@

                    Arg

                    Examples

                    - 
                    b = as.disk.frame(data.frame(a = 51:150, b = 1:100))
-d = as.disk.frame(data.frame(a = 151:250, b = 1:100))
-bd.df = merge(b, d, by = "b", merge_by_chunk_id = TRUE)
+    
                    b = as.disk.frame(data.frame(a = 51:150, b = 1:100))
+d = as.disk.frame(data.frame(a = 151:250, b = 1:100))
+bd.df = merge(b, d, by = "b", merge_by_chunk_id = TRUE)
 
 # clean up cars.df
-delete(b)
-delete(d)
-delete(bd.df)
                    
+delete(b)
+delete(d)
+delete(bd.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/move_to.html b/docs/reference/move_to.html index 4755299f..50090a6c 100644 --- a/docs/reference/move_to.html +++ b/docs/reference/move_to.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,9 +165,9 @@

                      Move or copy a disk.frame to another location

                      Move or copy a disk.frame to another location

                      - 
                      move_to(df, outdir, ..., copy = FALSE)
+    
                      move_to(df, outdir, ..., copy = FALSE)
 
-copy_df_to(df, outdir, ...)
                      
+copy_df_to(df, outdir, ...)

                      Arguments

                    • @@ -192,15 +195,16 @@

                    Value

                    a disk.frame

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
-cars_copy.df = copy_df_to(cars.df, outdir = tempfile(fileext=".df"))
+cars_copy.df = copy_df_to(cars.df, outdir = tempfile(fileext=".df"))
 
-cars2.df = move_to(cars.df, outdir = tempfile(fileext=".df"))
+cars2.df = move_to(cars.df, outdir = tempfile(fileext=".df"))
 
 # clean up
-delete(cars_copy.df)
-delete(cars2.df)
                    
+delete(cars_copy.df)
+delete(cars2.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/nchunks.html b/docs/reference/nchunks.html index 19440550..91ce528c 100644 --- a/docs/reference/nchunks.html +++ b/docs/reference/nchunks.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,15 +165,15 @@

                      Returns the number of chunks in a disk.frame

                      Returns the number of chunks in a disk.frame

                      - 
                      nchunks(df, ...)
+    
                      nchunks(df, ...)
 
-nchunk(df, ...)
+nchunk(df, ...)
 
 # S3 method for disk.frame
-nchunk(df, ...)
+nchunk(df, ...)
 
 # S3 method for disk.frame
-nchunks(df, skip.ready.check = FALSE, ...)
                      
+nchunks(df, skip.ready.check = FALSE, ...)

                      Arguments

                    • @@ -191,12 +194,15 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # return the number of chunks
-nchunks(cars.df)
                    #> [1] 6
                    nchunk(cars.df)
                    #> [1] 6
                    
+nchunks(cars.df)
+
                    #> [1] 6
                    nchunk(cars.df)
+
                    #> [1] 6
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/ncol_nrow.html b/docs/reference/ncol_nrow.html index 3b071e74..5d39a4fa 100644 --- a/docs/reference/ncol_nrow.html +++ b/docs/reference/ncol_nrow.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,15 +165,15 @@

                      Number of rows or columns

                      Number of rows or columns

                      - 
                      nrow(df, ...)
+    
                      nrow(df, ...)
 
 # S3 method for disk.frame
-nrow(df, ...)
+nrow(df, ...)
 
-ncol(df)
+ncol(df)
 
 # S3 method for disk.frame
-ncol(df)
                      
+ncol(df)

                      Arguments

                    • @@ -187,12 +190,15 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # return total number of column and rows
-ncol(cars.df)
                    #> [1] 2
                    nrow(cars.df)
                    #> [1] 50
                    
+ncol(cars.df)
+
                    #> [1] 2
                    nrow(cars.df)
+
                    #> [1] 50
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/one-stage-group-by-verbs.html b/docs/reference/one-stage-group-by-verbs.html index 5aa4d240..ec4a929e 100644 --- a/docs/reference/one-stage-group-by-verbs.html +++ b/docs/reference/one-stage-group-by-verbs.html @@ -74,7 +74,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -82,7 +82,7 @@
                  • @@ -138,7 +141,7 @@
                    • - +
                      • @@ -166,61 +169,61 @@

                      One Stage function

                      mean collected_agg

                      - 
                      var_df.chunk_agg.disk.frame(x, na.rm = FALSE)
+    
                      var_df.chunk_agg.disk.frame(x, na.rm = FALSE)
 
-var_df.collected_agg.disk.frame(listx)
+var_df.collected_agg.disk.frame(listx)
 
-sd_df.chunk_agg.disk.frame(x, na.rm = FALSE)
+sd_df.chunk_agg.disk.frame(x, na.rm = FALSE)
 
-sd_df.collected_agg.disk.frame(listx)
+sd_df.collected_agg.disk.frame(listx)
 
-mean_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
+mean_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
 
-mean_df.collected_agg.disk.frame(listx)
+mean_df.collected_agg.disk.frame(listx)
 
-sum_df.chunk_agg.disk.frame(x, ...)
+sum_df.chunk_agg.disk.frame(x, ...)
 
-sum_df.collected_agg.disk.frame(listx, ...)
+sum_df.collected_agg.disk.frame(listx, ...)
 
-min_df.chunk_agg.disk.frame(x, ...)
+min_df.chunk_agg.disk.frame(x, ...)
 
-min_df.collected_agg.disk.frame(listx, ...)
+min_df.collected_agg.disk.frame(listx, ...)
 
-max_df.chunk_agg.disk.frame(x, ...)
+max_df.chunk_agg.disk.frame(x, ...)
 
-max_df.collected_agg.disk.frame(listx, ...)
+max_df.collected_agg.disk.frame(listx, ...)
 
-median_df.chunk_agg.disk.frame(x, ...)
+median_df.chunk_agg.disk.frame(x, ...)
 
-median_df.collected_agg.disk.frame(listx, ...)
+median_df.collected_agg.disk.frame(listx, ...)
 
-n_df.chunk_agg.disk.frame(...)
+n_df.chunk_agg.disk.frame(...)
 
-n_df.collected_agg.disk.frame(listx, ...)
+n_df.collected_agg.disk.frame(listx, ...)
 
-length_df.chunk_agg.disk.frame(x, ...)
+length_df.chunk_agg.disk.frame(x, ...)
 
-length_df.collected_agg.disk.frame(listx, ...)
+length_df.collected_agg.disk.frame(listx, ...)
 
-any_df.chunk_agg.disk.frame(x, ...)
+any_df.chunk_agg.disk.frame(x, ...)
 
-any_df.collected_agg.disk.frame(listx, ...)
+any_df.collected_agg.disk.frame(listx, ...)
 
-all_df.chunk_agg.disk.frame(x, ...)
+all_df.chunk_agg.disk.frame(x, ...)
 
-all_df.collected_agg.disk.frame(listx, ...)
+all_df.collected_agg.disk.frame(listx, ...)
 
-n_distinct_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
+n_distinct_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
 
-n_distinct_df.collected_agg.disk.frame(listx, ...)
+n_distinct_df.collected_agg.disk.frame(listx, ...)
 
-quantile_df.chunk_agg.disk.frame(x, ...)
+quantile_df.chunk_agg.disk.frame(x, ...)
 
-quantile_df.collected_agg.disk.frame(listx, ...)
+quantile_df.collected_agg.disk.frame(listx, ...)
 
-IQR_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
+IQR_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
 
-IQR_df.collected_agg.disk.frame(listx, ...)
                      
+IQR_df.collected_agg.disk.frame(listx, ...)

                      Arguments

                    • @@ -259,7 +262,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/overwrite_check.html b/docs/reference/overwrite_check.html index 235c9466..386c7661 100644 --- a/docs/reference/overwrite_check.html +++ b/docs/reference/overwrite_check.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Check if the outdir exists or not

                      If the overwrite is TRUE then the folder will be deleted, otherwise the folder will be created.

                      - 
                      overwrite_check(outdir, overwrite)
                      + 
                      overwrite_check(outdir, overwrite)

                      Arguments

                    • @@ -179,12 +182,13 @@

                    Arg

                    Examples

                    - 
                    tf = tempfile()
-overwrite_check(tf, overwrite = FALSE)
-overwrite_check(tf, overwrite = TRUE)
+    
                    tf = tempfile()
+overwrite_check(tf, overwrite = FALSE)
+overwrite_check(tf, overwrite = TRUE)
 
 # clean up
-fs::dir_delete(tf)
                    
+fs::dir_delete(tf)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/print.disk.frame.html b/docs/reference/print.disk.frame.html index b8402b06..b3752bd0 100644 --- a/docs/reference/print.disk.frame.html +++ b/docs/reference/print.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,7 +166,7 @@

                      Print disk.frame

                      # S3 method for disk.frame
-print(x, ...)
                      +print(x, ...)

                      Arguments

                    • @@ -194,7 +197,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/pull.disk.frame.html b/docs/reference/pull.disk.frame.html index e22e0329..4bcc7b8a 100644 --- a/docs/reference/pull.disk.frame.html +++ b/docs/reference/pull.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,7 +166,7 @@

                      Pull a column from table similar to `dplyr::pull`.

                      # S3 method for disk.frame
-pull(.data, var = -1, name = NULL, ...)
                      +pull(.data, var = -1, name = NULL, ...)

                      Arguments

                    • @@ -202,7 +205,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/rbindlist.disk.frame.html b/docs/reference/rbindlist.disk.frame.html index 9226c689..2ea44cfb 100644 --- a/docs/reference/rbindlist.disk.frame.html +++ b/docs/reference/rbindlist.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,15 +165,15 @@

                      rbindlist disk.frames together

                      rbindlist disk.frames together

                      - 
                      rbindlist.disk.frame(
-  df_list,
-  outdir = tempfile(fileext = ".df"),
-  by_chunk_id = TRUE,
-  parallel = TRUE,
-  compress = 50,
-  overwrite = TRUE,
-  .progress = TRUE
-)
                      + 
                      rbindlist.disk.frame(
+  df_list,
+  outdir = tempfile(fileext = ".df"),
+  by_chunk_id = TRUE,
+  parallel = TRUE,
+  compress = 50,
+  overwrite = TRUE,
+  .progress = TRUE
+)

                      Arguments

                    • @@ -207,13 +210,15 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # row-bind two disk.frames
-cars2.df = rbindlist.disk.frame(list(cars.df, cars.df))
                    #> Appending disk.frames: 
                    
+cars2.df = rbindlist.disk.frame(list(cars.df, cars.df))
+
                    #> Appending disk.frames: 
                    
 # clean up cars.df
-delete(cars.df)
-delete(cars2.df)
                    
+delete(cars.df)
+delete(cars2.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/rechunk.html b/docs/reference/rechunk.html index 2abbba00..fb5df9b6 100644 --- a/docs/reference/rechunk.html +++ b/docs/reference/rechunk.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,16 +165,16 @@

                      Increase or decrease the number of chunks in the disk.frame

                      Increase or decrease the number of chunks in the disk.frame

                      - 
                      rechunk(
-  df,
-  nchunks,
-  outdir = attr(df, "path", exact = TRUE),
-  shardby = NULL,
-  overwrite = TRUE,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL
-)
                      + 
                      rechunk(
+  df,
+  nchunks,
+  outdir = attr(df, "path", exact = TRUE),
+  shardby = NULL,
+  overwrite = TRUE,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL
+)

                      Arguments

                    • @@ -213,21 +216,24 @@

                    Arg

                    Examples

                    # create a disk.frame with 2 chunks in tempdir()
-cars.df = as.disk.frame(cars, nchunks = 2)
+cars.df = as.disk.frame(cars, nchunks = 2)
 
 # re-chunking cars.df to 3 chunks, done "in-place" to the same folder as cars.df
-rechunk(cars.df, 3)
                    #> files have been backed up to temporary dir C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK\back_up_tmp_dir187c185718b5. You can recover there files until you restart your R session
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK\file187c36421ee.df"
+rechunk(cars.df, 3)
+
                    #> files have been backed up to temporary dir C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\back_up_tmp_dir4b144a1a2eda. You can recover there files until you restart your R session
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1466eb2237.df"
 #> nchunks: 3
 #> nrow (at source): 50
 #> ncol (at source): 2
 #> nrow (post operations): ???
 #> ncol (post operations): ???
                    
-new_path = tempfile(fileext = ".df")
+new_path = tempfile(fileext = ".df")
 # re-chunking cars.df to 4 chunks, shard by speed, and done "out-of-place" to a new directory
-cars2.df = rechunk(cars.df, 4, outdir=new_path, shardby = "speed")
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
+cars2.df = rechunk(cars.df, 4, outdir=new_path, shardby = "speed")
+
                    #> Hashing...
                    #> Hashing...
                    #> Hashing...
                    #> Appending disk.frames: 
                    
 # clean up cars.df
-delete(cars.df)
-delete(cars2.df)
                    +delete(cars.df) +delete(cars2.df) +
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/recommend_nchunks.html b/docs/reference/recommend_nchunks.html index f0c9e401..1b43a82d 100644 --- a/docs/reference/recommend_nchunks.html +++ b/docs/reference/recommend_nchunks.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                  • @@ -137,7 +140,7 @@
                    • - +
                      • @@ -164,13 +167,13 @@

                      Recommend number of chunks based on input size

                      into. It can accept filesizes in bytes (as integer) or a data.frame

                      - 
                      recommend_nchunks(
-  df,
-  type = "csv",
-  minchunks = data.table::getDTthreads(),
-  conservatism = 8,
-  ram_size = df_ram_size()
-)
                      + 
                      recommend_nchunks(
+  df,
+  type = "csv",
+  minchunks = data.table::getDTthreads(),
+  conservatism = 8,
+  ram_size = df_ram_size()
+)

                      Arguments

                    • @@ -205,9 +208,11 @@

                    Arg

                    Examples

                    # recommend nchunks based on data.frame
-recommend_nchunks(cars)
                    #> [1] 6
                    
+recommend_nchunks(cars)
+
                    #> [1] 6
                    
 # recommend nchunks based on file size ONLY CSV is implemented at the moment
-recommend_nchunks(1024^3)
                    #> [1] 6
                    +recommend_nchunks(1024^3) +
                    #> [1] 6
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/remove_chunk.html b/docs/reference/remove_chunk.html index 00048845..6e5db636 100644 --- a/docs/reference/remove_chunk.html +++ b/docs/reference/remove_chunk.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Removes a chunk from the disk.frame

                      Removes a chunk from the disk.frame

                      - 
                      remove_chunk(df, chunk_id, full.names = FALSE)
                      + 
                      remove_chunk(df, chunk_id, full.names = FALSE)

                      Arguments

                    • @@ -184,31 +187,38 @@

                    Arg

                    Examples

                    # TODO add these to tests
-cars.df = as.disk.frame(cars, nchunks = 4)
+cars.df = as.disk.frame(cars, nchunks = 4)
 
 # removes 3rd chunk
-remove_chunk(cars.df, 3)
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK\file187c1fad1470.df"
+remove_chunk(cars.df, 3)
+
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1473bc5984.df"
 #> nchunks: 3
 #> nrow (at source): 37
 #> ncol (at source): 2
 #> nrow (post operations): ???
-#> ncol (post operations): ???
                    nchunks(cars.df) # 3
                    #> [1] 3
                    
+#> ncol (post operations): ???
                    nchunks(cars.df) # 3
+
                    #> [1] 3
                    
 # removes 4th chunk
-remove_chunk(cars.df, "4.fst")
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK\file187c1fad1470.df"
+remove_chunk(cars.df, "4.fst")
+
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1473bc5984.df"
 #> nchunks: 2
 #> nrow (at source): 26
 #> ncol (at source): 2
 #> nrow (post operations): ???
-#> ncol (post operations): ???
                    nchunks(cars.df) # 3
                    #> [1] 2
                    
+#> ncol (post operations): ???
                    nchunks(cars.df) # 3
+
                    #> [1] 2
                    
 # removes 2nd chunk
-remove_chunk(cars.df, file.path(attr(cars.df, "path", exact=TRUE), "2.fst"), full.names = TRUE)
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK\file187c1fad1470.df"
+remove_chunk(cars.df, file.path(attr(cars.df, "path", exact=TRUE), "2.fst"), full.names = TRUE)
+
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1473bc5984.df"
 #> nchunks: 1
 #> nrow (at source): 13
 #> ncol (at source): 2
 #> nrow (post operations): ???
-#> ncol (post operations): ???
                    nchunks(cars.df) # 1
                    #> [1] 1
                    
+#> ncol (post operations): ???
                    nchunks(cars.df) # 1
+
                    #> [1] 1
                    
 # clean up cars.df
-delete(cars.df)
                    +delete(cars.df) +
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/sample.html b/docs/reference/sample.html index 161546d8..83e4f100 100644 --- a/docs/reference/sample.html +++ b/docs/reference/sample.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -163,7 +166,7 @@

                      Sample n rows from a disk.frame

                      # S3 method for disk.frame
-sample_frac(tbl, size = 1, replace = FALSE, weight = NULL, .env = NULL, ...)
                      +sample_frac(tbl, size = 1, replace = FALSE, weight = NULL, .env = NULL, ...)

                      Arguments

                    • @@ -175,8 +178,8 @@

                    Arg

                    @@ -201,33 +204,36 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
-
-collect(sample_frac(cars.df, 0.5))
                    #>    speed dist
-#> 1     10   34
-#> 2      8   16
-#> 3      4   10
-#> 4     10   26
-#> 5     12   14
-#> 6     13   26
-#> 7     11   28
-#> 8     13   34
-#> 9     13   46
-#> 10    15   26
-#> 11    16   32
-#> 12    15   54
-#> 13    18   42
-#> 14    17   32
-#> 15    18   84
-#> 16    17   40
-#> 17    22   66
-#> 18    19   68
-#> 19    20   48
-#> 20    20   52
-#> 21    24  120
-#> 22    25   85
                    
+    
                    cars.df = as.disk.frame(cars)
+
+collect(sample_frac(cars.df, 0.5))
+
                    #>     speed dist
+#>  1:    10   34
+#>  2:     7   22
+#>  3:     8   16
+#>  4:     4    2
+#>  5:    11   17
+#>  6:    12   20
+#>  7:    11   28
+#>  8:    13   34
+#>  9:    15   26
+#> 10:    15   54
+#> 11:    15   20
+#> 12:    14   80
+#> 13:    16   40
+#> 14:    18   42
+#> 15:    17   40
+#> 16:    19   36
+#> 17:    19   68
+#> 18:    20   64
+#> 19:    20   52
+#> 20:    20   48
+#> 21:    25   85
+#> 22:    24  120
+#>     speed dist
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/setup_disk.frame.html b/docs/reference/setup_disk.frame.html index 2c8a7cda..477f1e08 100644 --- a/docs/reference/setup_disk.frame.html +++ b/docs/reference/setup_disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,12 +165,12 @@

                      Set up disk.frame environment

                      Set up disk.frame environment

                      - 
                      setup_disk.frame(
-  workers = data.table::getDTthreads(),
-  future_backend = future::multisession,
-  ...,
-  gui = FALSE
-)
                      + 
                      setup_disk.frame(
+  workers = data.table::getDTthreads(),
+  future_backend = future::multisession,
+  ...,
+  gui = FALSE
+)

                      Arguments

                    • size

                    <tidy-select> -For sample_n(), the number of rows to select. -For sample_frac(), the fraction of rows to select. +For sample_n(), the number of rows to select. +For sample_frac(), the fraction of rows to select. If tbl is grouped, size applies to each group.
                    @@ -192,21 +195,23 @@

                    Arg

                    Examples

                    - 
                    if (interactive()) {
+    
                    if (interactive()) {
   # setup disk.frame to use multiple workers these may use more than two
   # cores, and is therefore not allowed on CRAN. Hence it's set to run only in
   # interactive session
-  setup_disk.frame()
-
+  setup_disk.frame()
+  
   # use a Shiny GUI to adjust settings
   # only run in interactive()
-  setup_disk.frame(gui = TRUE)
-}
+  setup_disk.frame(gui = TRUE)
+}
 
 # set the number workers to 2
-setup_disk.frame(2)
                    #> The number of workers available for disk.frame is 2
                    
+setup_disk.frame(2)
+
                    #> The number of workers available for disk.frame is 2
                    
 # if you do not wish to use multiple workers you can set it to sequential
-setup_disk.frame(future_backend=future::sequential)
                    #> Warning: Ignored 2 unknown arguments: 'workers', 'gc'
                    #> The number of workers available for disk.frame is 1
                    
+setup_disk.frame(future_backend=future::sequential)
+
                    #> Warning: Detected 1 unknown future arguments: 'workers'
                    #> The number of workers available for disk.frame is 1
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/shard.html b/docs/reference/shard.html index 13f9501d..6e4ed76b 100644 --- a/docs/reference/shard.html +++ b/docs/reference/shard.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                  • @@ -137,7 +140,7 @@
                    • - +
                      • @@ -164,19 +167,19 @@

                      Shard a data.frame/data.table or disk.frame into chunk and saves it into a d

                      `distribute` is an alias for `shard`

                      - 
                      shard(
-  df,
-  shardby,
-  outdir = tempfile(fileext = ".df"),
-  ...,
-  nchunks = recommend_nchunks(df),
-  overwrite = FALSE,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL
-)
+    
                      shard(
+  df,
+  shardby,
+  outdir = tempfile(fileext = ".df"),
+  ...,
+  nchunks = recommend_nchunks(df),
+  overwrite = FALSE,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL
+)
 
-distribute(...)
                      
+distribute(...)

                      Arguments

                    • @@ -223,9 +226,11 @@

                    Arg

                    Examples

                    
 # shard the cars data.frame by speed so that rows with the same speed are in the same chunk
-iris.df = shard(iris, "Species")
                    #> Hashing...
                    
+iris.df = shard(iris, "Species")
+
                    #> Hashing...
                    
 # clean up cars.df
-delete(iris.df)
                    +delete(iris.df) +
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/shardkey.html b/docs/reference/shardkey.html index 4e99bbca..a23b17ef 100644 --- a/docs/reference/shardkey.html +++ b/docs/reference/shardkey.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Returns the shardkey (not implemented yet)

                      Returns the shardkey (not implemented yet)

                      - 
                      shardkey(df)
                      + 
                      shardkey(df)

                      Arguments

                    • @@ -189,7 +192,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/shardkey_equal.html b/docs/reference/shardkey_equal.html index 5c4e1ad8..14596490 100644 --- a/docs/reference/shardkey_equal.html +++ b/docs/reference/shardkey_equal.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,7 +165,7 @@

                      Compare two disk.frame shardkeys

                      Compare two disk.frame shardkeys

                      - 
                      shardkey_equal(sk1, sk2)
                      + 
                      shardkey_equal(sk1, sk2)

                      Arguments

                    • @@ -193,7 +196,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/show_ceremony.html b/docs/reference/show_ceremony.html index 7481dc90..f3db3865 100644 --- a/docs/reference/show_ceremony.html +++ b/docs/reference/show_ceremony.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -162,13 +165,13 @@

                      Show the code to setup disk.frame

                      Show the code to setup disk.frame

                      - 
                      show_ceremony()
+    
                      show_ceremony()
 
-ceremony_text()
+ceremony_text()
 
-show_boilerplate()
+show_boilerplate()
 
-insert_ceremony()
                      
+insert_ceremony()
                      @@ -187,7 +190,7 @@

                      Contents

                      -

                      Site built with pkgdown 1.5.1.

                      +

                      Site built with pkgdown 1.6.1.

                      diff --git a/docs/reference/srckeep.html b/docs/reference/srckeep.html index fe997fbd..898b76fa 100644 --- a/docs/reference/srckeep.html +++ b/docs/reference/srckeep.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                    • @@ -136,7 +139,7 @@
                      • - +
                        • @@ -162,9 +165,9 @@

                        Keep only the variables from the input listed in selections

                        Keep only the variables from the input listed in selections

                        - 
                        srckeep(diskf, selections, ...)
+    
                        srckeep(diskf, selections, ...)
 
-srckeepchunks(diskf, chunks, ...)
                        
+srckeepchunks(diskf, chunks, ...)

                        Arguments

                    • @@ -189,10 +192,11 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # when loading cars's chunks into RAM, load only the column speed
-collect(srckeep(cars.df, "speed"))
                    #>     speed
+collect(srckeep(cars.df, "speed"))
+
                    #>     speed
 #>  1:     4
 #>  2:     4
 #>  3:     7
@@ -245,7 +249,8 @@ 
                    Examp
 #> 50:    25
 #>     speed
                    
 # clean up cars.df
-delete(cars.df)
                    
+delete(cars.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/sub-.disk.frame.html b/docs/reference/sub-.disk.frame.html index 961cbbe9..40e28919 100644 --- a/docs/reference/sub-.disk.frame.html +++ b/docs/reference/sub-.disk.frame.html @@ -72,7 +72,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -80,7 +80,7 @@
                  • @@ -136,7 +139,7 @@
                    • - +
                      • @@ -208,11 +211,13 @@

                      Arg

                      Examples

                      - 
                      cars.df = as.disk.frame(cars)
-speed_limit = 50
-cars.df[speed < speed_limit ,.N, cut(dist, pretty(dist))]
                      #> Error in .checkTypos(e, names_x): Object 'speed_limit' not found amongst speed, dist
                      
+    
                      cars.df = as.disk.frame(cars)
+speed_limit = 50
+cars.df[speed < speed_limit ,.N, cut(dist, pretty(dist))]
+
                      #> Error in .checkTypos(e, names_x): Object 'speed_limit' not found amongst speed, dist
                      
 # clean up
-delete(cars.df)
                      
+delete(cars.df)
+
                      -

                      Site built with pkgdown 1.5.1.

                      +

                      Site built with pkgdown 1.6.1.

                      diff --git a/docs/reference/tbl_vars.disk.frame.html b/docs/reference/tbl_vars.disk.frame.html index f7d0febf..4869fa2f 100644 --- a/docs/reference/tbl_vars.disk.frame.html +++ b/docs/reference/tbl_vars.disk.frame.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                    • @@ -137,7 +140,7 @@
                      • - +
                        • @@ -165,10 +168,10 @@

                        Column names for RStudio auto-complete

                        # S3 method for disk.frame
-tbl_vars(x)
+tbl_vars(x)
 
 # S3 method for disk.frame
-group_vars(x)
                        +group_vars(x)

                        Arguments

                    • @@ -195,7 +198,7 @@

                    Contents

                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/write_disk.frame.html b/docs/reference/write_disk.frame.html index b8e57f50..34f8bdf9 100644 --- a/docs/reference/write_disk.frame.html +++ b/docs/reference/write_disk.frame.html @@ -73,7 +73,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -81,7 +81,7 @@
                  • @@ -137,7 +140,7 @@
                    • - +
                      • @@ -164,21 +167,21 @@

                      Write disk.frame to disk

                      then using the as.disk.frame function is recommended for most cases

                      - 
                      write_disk.frame(
-  df,
-  outdir = tempfile(fileext = ".df"),
-  nchunks = ifelse("disk.frame" %in% class(df), nchunks.disk.frame(df),
-    recommend_nchunks(df)),
-  overwrite = FALSE,
-  shardby = NULL,
-  compress = 50,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL,
-  ...
-)
-
-output_disk.frame(...)
                      + 
                      write_disk.frame(
+  df,
+  outdir = tempfile(fileext = ".df"),
+  nchunks = ifelse("disk.frame" %in% class(df), nchunks.disk.frame(df),
+    recommend_nchunks(df)),
+  overwrite = FALSE,
+  shardby = NULL,
+  compress = 50,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL,
+  ...
+)
+
+output_disk.frame(...)

                      Arguments

                    • @@ -227,11 +230,12 @@

                    Arg

                    Examples

                    - 
                    cars.df = as.disk.frame(cars)
+    
                    cars.df = as.disk.frame(cars)
 
 # write out a lazy disk.frame to disk
-cars2.df = write_disk.frame(cmap(cars.df, ~.x[1,]), overwrite = TRUE)
-collect(cars2.df)
                    #>    speed dist
+cars2.df = write_disk.frame(cmap(cars.df, ~.x[1,]), overwrite = TRUE)
+collect(cars2.df)
+
                    #>    speed dist
 #> 1:     4    2
 #> 2:    11   17
 #> 3:    13   46
@@ -239,8 +243,9 @@ 
                    Examp
 #> 5:    19   46
 #> 6:    24   70
                    
 # clean up cars.df
-delete(cars.df)
-delete(cars2.df)
                    
+delete(cars.df)
+delete(cars2.df)
+
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/docs/reference/zip_to_disk.frame.html b/docs/reference/zip_to_disk.frame.html index 862dab16..a6f1a16e 100644 --- a/docs/reference/zip_to_disk.frame.html +++ b/docs/reference/zip_to_disk.frame.html @@ -75,7 +75,7 @@ disk.frame - 0.3.6 + 0.5.0 @@ -83,7 +83,7 @@
                  • @@ -139,7 +142,7 @@
                    • - +
                      • @@ -167,13 +170,13 @@

                      `zip_to_disk.frame` is used to read and convert every CSV file within the zi file to disk.frame format

                      - 
                      zip_to_disk.frame(
-  zipfile,
-  outdir,
-  ...,
-  validation.check = FALSE,
-  overwrite = TRUE
-)
                      + 
                      zip_to_disk.frame(
+  zipfile,
+  outdir,
+  ...,
+  validation.check = FALSE,
+  overwrite = TRUE
+)

                      Arguments

                    • @@ -210,25 +213,27 @@

                    See a

                    Examples

                    # create a zip file containing a csv
-csvfile = tempfile(fileext = ".csv")
-write.csv(cars, csvfile)
-zipfile = tempfile(fileext = ".zip")
-zip(zipfile, csvfile)
+csvfile = tempfile(fileext = ".csv")
+write.csv(cars, csvfile)
+zipfile = tempfile(fileext = ".zip")
+zip(zipfile, csvfile)
 
 # read every file and convert it to a disk.frame
-zip.df = zip_to_disk.frame(zipfile, tempfile(fileext = ".df"))
+zip.df = zip_to_disk.frame(zipfile, tempfile(fileext = ".df"))
 
 # there is only one csv file so it return a list of one disk.frame
-zip.df[[1]]
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\RtmpInritK\file187c1d9666ec.df/Users/RTX2080/AppData/Local/Temp/RtmpInritK/file187c725c4b4.csv"
+zip.df[[1]]
+
                    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b14559e63cf.df/Users/RTX2080/AppData/Local/Temp/Rtmp2PGAkP/file4b1437ba44b8.csv"
 #> nchunks: 6
 #> nrow (at source): 50
 #> ncol (at source): 3
 #> nrow (post operations): ???
 #> ncol (post operations): ???
                    
 # clean up
-unlink(csvfile)
-unlink(zipfile)
-delete(zip.df[[1]])
                    +unlink(csvfile) +unlink(zipfile) +delete(zip.df[[1]]) +
                    -

                    Site built with pkgdown 1.5.1.

                    +

                    Site built with pkgdown 1.6.1.

                    diff --git a/man/dplyr_verbs.Rd b/man/dplyr_verbs.Rd index d3930c57..340e6958 100644 --- a/man/dplyr_verbs.Rd +++ b/man/dplyr_verbs.Rd @@ -8,7 +8,6 @@ \alias{transmute.disk.frame} \alias{arrange.disk.frame} \alias{chunk_arrange} -\alias{add_count.disk.frame} \alias{add_tally.disk.frame} \alias{do.disk.frame} \alias{distinct.disk.frame} @@ -30,8 +29,6 @@ chunk_arrange(.data, ...) -add_count.disk.frame(.data, ...) - add_tally.disk.frame(.data, ...) \method{do}{disk.frame}(.data, ...) diff --git a/tests/testthat.R b/tests/testthat.R deleted file mode 100644 index 6681ff05..00000000 --- a/tests/testthat.R +++ /dev/null @@ -1,4 +0,0 @@ -library(testthat) -library(disk.frame) - -test_check("disk.frame") diff --git a/tests/testthat/test-Rcpp.R b/tests/testthat/test-Rcpp.R deleted file mode 100644 index d4144ca9..00000000 --- a/tests/testthat/test-Rcpp.R +++ /dev/null @@ -1,6 +0,0 @@ -context("test-RcppExprts") - - -test_that("testing Rccpexports nothing here", { - expect_equal(2L, 2L) -}) \ No newline at end of file diff --git a/tests/testthat/test-add-chunk.r b/tests/testthat/test-add-chunk.r deleted file mode 100644 index c6b38a73..00000000 --- a/tests/testthat/test-add-chunk.r +++ /dev/null @@ -1,65 +0,0 @@ -context("test-add-chunk") - -setup({ - setup_disk.frame(workers = 2) -}) - -test_that("guard against github 292", { - a = data.frame(a = as.Date("2020-07-01"), b = runif(1e6)) - - a.df = as.disk.frame(a) - - head(a.df) - - expect_s3_class(add_chunk(a.df, a), "disk.frame") - delete(a.df) -}) - -test_that("testing add chunk without naming chunk_id", { - a = data.frame(a = 1:100, b = 1:100) - - a1 = as.disk.frame(a, overwrite = TRUE) - - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - add_chunk(a1, b) - expect_equal(nrow(a1), 200) - - add_chunk(a1, d) - expect_equal(nrow(a1), 250) - - delete(a1) -}) - -test_that("testing add chunk by naming chunk_id", { - a = data.frame(a = 1:100, b = 1:100) - - a1 = as.disk.frame(a, overwrite = TRUE) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - add_chunk(a1, b, chunk_id = nchunks(a1)+2) - expect_equal(nrow(a1), 200) - - add_chunk(a1, d, chunk_id = nchunks(a1)+2) - expect_equal(nrow(a1), 250) - - delete(a1) -}) - -test_that("testing add chunk by using compression", { - a = data.frame(a = 1:100, b = 1:100) - - a1 = as.disk.frame(a, overwrite = TRUE) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - add_chunk(a1, b, compress=50) - expect_equal(nrow(a1), 200) - - delete(a1) -}) - -teardown({ -}) \ No newline at end of file diff --git a/tests/testthat/test-anti_join.R b/tests/testthat/test-anti_join.R deleted file mode 100644 index 348ebaf2..00000000 --- a/tests/testthat/test-anti_join.R +++ /dev/null @@ -1,75 +0,0 @@ -context("test-anti_join") - -setup({ - setup_disk.frame(workers = 2) - - a = data.frame(a = 1:100, b = 1:100) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_aj.df"), nchunks = 4, overwrite = TRUE) - as.disk.frame(b, file.path(tempdir(), "tmp_b_aj.df"), nchunks = 5, overwrite = TRUE) - as.disk.frame(d, file.path(tempdir(), "tmp_d_aj.df"), overwrite = TRUE) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_aj2.df"), nchunks = 4, overwrite = TRUE) - as.disk.frame(b, file.path(tempdir(), "tmp_b_aj2.df"), nchunks = 5, overwrite = TRUE) - as.disk.frame(d, file.path(tempdir(), "tmp_d_aj2.df"), overwrite = TRUE) -}) - -test_that("testing anti_join where right is data.frame", { - #skip_on_cran() - a = disk.frame(file.path(tempdir(), "tmp_a_aj.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_aj.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_aj.df")) - bc = collect(b) - dc = collect(d) - - abc = anti_join(a, bc, by = "a") %>% collect - expect_equal(nrow(abc), 50) - - abc0 = anti_join(a, bc, by = c("a","b")) %>% collect - expect_equal(nrow(abc0), 100) - - abc100 = anti_join(a, bc, by = "b") %>% collect - expect_equal(nrow(abc100), 0) - - abd50 = anti_join(a, dc, by = "b") %>% collect - expect_equal(nrow(abd50), 50) -}) - -test_that("testing anti_join where right is disk.frame", { - #skip_on_cran() - a = disk.frame(file.path(tempdir(),"tmp_a_aj2.df")) - b = disk.frame(file.path(tempdir(),"tmp_b_aj2.df")) - d = disk.frame(file.path(tempdir(),"tmp_d_aj2.df")) - - expect_warning({ - ab <- anti_join(a, b, by = "a", merge_by_chunk_id = FALSE) %>% collect - }) - expect_equal(nrow(ab), 50) - - expect_warning({ - ab0 = anti_join(a, b, by = c("a","b"), merge_by_chunk_id = FALSE) %>% collect - }) - expect_equal(nrow(ab0), 100) - - expect_warning({ - ab100 = anti_join(a, b, by = "b", merge_by_chunk_id = FALSE) %>% collect - }) - expect_equal(nrow(ab100), 0) - - expect_warning({ - ad50 = anti_join(a, d, by = "b", merge_by_chunk_id = FALSE) %>% collect - }) - expect_equal(nrow(ad50), 50) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(),"tmp_a_aj.df")) - fs::dir_delete(file.path(tempdir(),"tmp_b_aj.df")) - fs::dir_delete(file.path(tempdir(),"tmp_d_aj.df")) - - fs::dir_delete(file.path(tempdir(),"tmp_a_aj2.df")) - fs::dir_delete(file.path(tempdir(),"tmp_b_aj2.df")) - fs::dir_delete(file.path(tempdir(),"tmp_d_aj2.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-as-data-frame.R b/tests/testthat/test-as-data-frame.R deleted file mode 100644 index 0cb853fb..00000000 --- a/tests/testthat/test-as-data-frame.R +++ /dev/null @@ -1,14 +0,0 @@ -context("test-as-data-frame") - -test_that("as.data.frame works", { - tmpdir = tempfile("disk.frame.tmp") - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11, 100), tmpdir, overwrite = T) - dff = as.data.frame(df) - dft = data.table::as.data.table(df) - expect_s3_class(dff, "data.frame") - expect_s3_class(dft, "data.table") - expect_equal(nrow(dff), 1e5+11) - - delete(df) -}) - diff --git a/tests/testthat/test-as-disk-frame.R b/tests/testthat/test-as-disk-frame.R deleted file mode 100644 index b1ac42dc..00000000 --- a/tests/testthat/test-as-disk-frame.R +++ /dev/null @@ -1,15 +0,0 @@ -context("test-as-disk-frame") - -test_that("as.disk.frame works", { - ROWS = 1e5+11 - - df = disk.frame:::gen_datatable_synthetic(ROWS) - tf = file.path(tempdir(), "tmp_as_disk_frame_delete") - - dfdf <- as.disk.frame(df, outdir = tf, overwrite=TRUE) - - expect_equal(nrow(dfdf), ROWS) - expect_error(dfdf <- as.disk.frame(df, tf, overwrite=FALSE)) - - delete(dfdf) -}) diff --git a/tests/testthat/test-bloom-filter.r b/tests/testthat/test-bloom-filter.r deleted file mode 100644 index 9fa4ad42..00000000 --- a/tests/testthat/test-bloom-filter.r +++ /dev/null @@ -1,16 +0,0 @@ -# context("test-bloomfilter") -# -# test_that("bloomfilter should fail here", { -# expect_error(make_bloomfilter(df, c("origin", "dest"))) -# }) -# -# test_that("bloomfilter should succeed", { -# df = nycflights13::flights %>% as.disk.frame(shardby = c("carrier")) -# make_bloomfilter(df, "carrier") -# expect_true(length(bf_likely_in_chunks(df, "carrier", "UA")) == 1) -# -# expect_equal(nrow(collect(use_bloom_filter(df, "carrier", "UA"))), nrow(filter(nycflights13::flights, carrier == "UA"))) -# -# # clean up -# delete(df) -# }) diff --git a/tests/testthat/test-collect.R b/tests/testthat/test-collect.R deleted file mode 100644 index b5a174c8..00000000 --- a/tests/testthat/test-collect.R +++ /dev/null @@ -1,57 +0,0 @@ -context("test-collect") - -setup({ - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11), file.path(tempdir(),"tmp_col_delete"), overwrite=T) -}) - -test_that("collect works on simple data", { - df = disk.frame(file.path(tempdir(),"tmp_col_delete")) - dff = dplyr::collect(df) - expect_equal(nrow(dff), 1e5+11) - expect_s3_class(dff, "data.frame") - expect_s3_class(dff, "data.table") -}) - -test_that("collect works on lazy stream", { - df = disk.frame(file.path(tempdir(),"tmp_col_delete")) - df = cmap(df, lazy = T, ~{ - .x[1:10, ] - }) - dff = dplyr::collect(df) - expect_equal(nrow(dff), nchunks(df)*10) - expect_s3_class(dff, "data.frame") - expect_s3_class(dff, "data.table") -}) - -test_that("collect works on lazy stream followed by dplyr", { - df = disk.frame(file.path(tempdir(),"tmp_col_delete")) - df = cmap(df, lazy = T, ~{ - .x[1:10, ] - }) %>% select(id1, id4) - - dff = dplyr::collect(df) - expect_equal(nrow(dff), nchunks(df)*10) - expect_equal(ncol(dff), 2) - expect_s3_class(dff, "data.frame") - expect_s3_class(dff, "data.table") -}) - - -test_that("collect works on dplyr::select followed by lazy", { - df = disk.frame(file.path(tempdir(),"tmp_col_delete")) - df = df %>% select(id1, id4) %>% - cmap.disk.frame(lazy = T, ~{ - .x[1:10, ] - }) - - dff = dplyr::collect(df) - expect_equal(nrow(dff), nchunks(df)*10) - expect_equal(ncol(dff), 2) - expect_s3_class(dff, "data.frame") - expect_s3_class(dff, "data.table") -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(),"tmp_col_delete")) -}) \ No newline at end of file diff --git a/tests/testthat/test-compute.r b/tests/testthat/test-compute.r deleted file mode 100644 index 3a1bd56e..00000000 --- a/tests/testthat/test-compute.r +++ /dev/null @@ -1,55 +0,0 @@ -context("test-compute") - -setup({ - setup_disk.frame(workers = 2) - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11), file.path(tempdir(),"tmp_compute_delete"), overwrite=T) -}) - -test_that("compute works on simple data", { - df = disk.frame(file.path(tempdir(),"tmp_compute_delete")) - dff = compute(df) - - expect_equal(nrow(dff), 1e5+11) - expect_s3_class(dff, "disk.frame") -}) - -test_that("compute works on lazy stream", { - df = disk.frame(file.path(tempdir(),"tmp_compute_delete")) - df = cmap(df, lazy = T, ~{ - .x[1:10, ] - }) - dff = compute(df) - expect_equal(nrow(dff), nchunks(df)*10) - expect_s3_class(dff, "disk.frame") -}) - -test_that("compute works on lazy stream followed by dplyr", { - df = disk.frame(file.path(tempdir(),"tmp_compute_delete")) - df = cmap(df, lazy = T, ~{ - .x[1:10, ] - }) %>% select(id1, id4) - - dff = compute(df) - expect_equal(nrow(dff), nchunks(df)*10) - expect_equal(ncol(dff), 2) - expect_s3_class(dff, "disk.frame") -}) - - -test_that("compute works on dplyr::select followed by lazy", { - df = disk.frame(file.path(tempdir(),"tmp_compute_delete")) - df = df %>% select(id1, id4) %>% - cmap(lazy = T, ~{ - .x[1:10, ] - }) - - dff = dplyr::collect(df) - expect_equal(nrow(dff), nchunks(df)*10) - expect_equal(ncol(dff), 2) - expect_s3_class(dff, "data.frame") -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(),"tmp_compute_delete")) -}) \ No newline at end of file diff --git a/tests/testthat/test-csv2disk.frame.r b/tests/testthat/test-csv2disk.frame.r deleted file mode 100644 index dcc73652..00000000 --- a/tests/testthat/test-csv2disk.frame.r +++ /dev/null @@ -1,84 +0,0 @@ -context("test-csv2disk.frame") - -setup({ - df = disk.frame:::gen_datatable_synthetic(1e3+11) - data.table::fwrite(df, file.path(tempdir(), "tmp_pls_delete_csv2df.csv")) - data.table::fwrite(df, file.path(tempdir(), "tmp_pls_delete_csv2df2.csv")) - data.table::fwrite(df, file.path(tempdir(), "tmp_pls_delete_csv2df3.csv")) -}) - -test_that("csv2disk.frame works with no shard", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_csv2df.csv"), - file.path(tempdir(), "tmp_pls_delete_csv2df.df"), - overwrite=TRUE, - nchunks=max(2, recommend_nchunks(file.size(file.path(tempdir(), "tmp_pls_delete_csv2df.csv"))))) - dff1 = dff[,sum(v1), id1] - dff2 = dff1[,sum(V1), id1] - expect_false(nrow(dff1) == nrow(dff2)) - expect_equal(nrow(dff), 1e3+11) - expect_equal(ncol(dff), 10) -}) - -test_that("csv2disk.frame works with shard", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_csv2df2.csv"), - file.path(tempdir(), "tmp_pls_delete_csv2df2.df"), - shardby = "id1", overwrite = TRUE) - dff1 = dff[,sum(v1), id1] - dff2 = dff1[,sum(V1), id1] - expect_true(nrow(dff1) == nrow(dff2)) - expect_equal(nrow(dff), 1e3+11) - expect_equal(ncol(dff), 10) - - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_csv2df3.csv"), - file.path(tempdir(), "tmp_pls_delete_csv2df3.df"), - shardby = c("id1","id2")) - - dff1 = dff[,sum(v1), .(id1,id2)] - dff2 = dff1[,sum(V1), .(id1,id2)] - expect_true(nrow(dff1) == nrow(dff2)) - expect_equal(nrow(dff), 1e3+11) - expect_equal(ncol(dff), 10) -}) - -test_that("csv2disk.frame tests readr", { - library(dplyr) - library(disk.frame) - library(data.table) - library(nycflights13) - - expect_equal(1,1) - - # TODO make this test better - # convert from a data frame - # flights <- flights %>% - # dplyr::mutate(date = as.Date(paste(year, month, day, sep = "-"))) - # str(flights) # time_hour is POSIXct - # - # flights.df <- as.disk.frame( - # flights, - # outdir = file.path(tempdir(), "tmp_flights.df"), - # overwrite = TRUE) - # flights.df - # str(collect(flights.df)) - # - # # with sharding - # df_path = file.path(tempdir(), "tmp_flights.df") - # flights.df <- csv_to_disk.frame( - # csv_path, - # outdir = df_path, - # shardby = "minute", - # overwrite = T, - # backend = "readr") -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_pls_delete_csv2df.df")) - fs::dir_delete(file.path(tempdir(), "tmp_pls_delete_csv2df2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_pls_delete_csv2df3.df")) - fs::file_delete(file.path(tempdir(), "tmp_pls_delete_csv2df.csv")) - fs::file_delete(file.path(tempdir(), "tmp_pls_delete_csv2df2.csv")) - fs::file_delete(file.path(tempdir(), "tmp_pls_delete_csv2df3.csv")) -}) \ No newline at end of file diff --git a/tests/testthat/test-data-table.r b/tests/testthat/test-data-table.r deleted file mode 100644 index 0275ee4a..00000000 --- a/tests/testthat/test-data-table.r +++ /dev/null @@ -1,41 +0,0 @@ -context("test-data.table [") - -setup({ - library(data.table) - setup_disk.frame(workers = 2) - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11), file.path(tempdir(), "tmp_col_delete"), overwrite=TRUE, nchunks = 8) -}) - -test_that("data.table .N", { - library(data.table) - df = disk.frame(file.path(tempdir(), "tmp_col_delete")) - res <- sum(unlist(df[,.N])) - expect_equal(res , 1e5+11) -}) - -test_that("data.table .N+y V1", { - df = disk.frame(file.path(tempdir(), "tmp_col_delete")) - if(interactive()) { - y = 2 - - {y = 3; a <- df[,.(n_plus_y = .N + y), v1]} - b <- df[,.N, v1] - - expect_equal(a$n_plus_y, b$N + y) - } else { - # TODO figure out why the above fails - expect_equal(2L, 2L) - } -}) - -test_that("data.table do not return a data.table", { - library(data.table) - df = disk.frame(file.path(tempdir(), "tmp_col_delete")) - res <- df[,.(.N), rbind=FALSE] - expect_equal(typeof(res), "list") - expect_equal(length(res), 8) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_col_delete")) -}) diff --git a/tests/testthat/test-delete.r b/tests/testthat/test-delete.r deleted file mode 100644 index cc13cce8..00000000 --- a/tests/testthat/test-delete.r +++ /dev/null @@ -1,20 +0,0 @@ -context("test-delete") - -setup({ - setup_disk.frame(workers = 2) - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11), file.path(tempdir(), "tmp_del_delete"), overwrite = TRUE) -}) - -test_that("data.table .N", { - df = disk.frame(file.path(tempdir(), "tmp_del_delete")) - p = attr(df, "path", exact=TRUE) - expect_true(fs::dir_exists(p)) - - delete(df) - - expect_false(fs::dir_exists(p)) -}) - -teardown({ - #fs::dir_delete("tmp_del_delete") -}) \ No newline at end of file diff --git a/tests/testthat/test-disk-frame.r b/tests/testthat/test-disk-frame.r deleted file mode 100644 index 236e7a18..00000000 --- a/tests/testthat/test-disk-frame.r +++ /dev/null @@ -1,34 +0,0 @@ -context("test-disk.frame") - -# TODO add in tests here - -test_that("test add_meta", { - # it works so how to test this? - expect_equal(2L, 2L) - -}) - -test_that("test head", { - # it works so how to test this? - expect_equal(2L, 2L) - -}) - -test_that("test tail", { - # it works so how to test this? - expect_equal(2L, 2L) - -}) - - -test_that("test nrow", { - # it works so how to test this? - - expect_equal(2L, 2L) -}) - -test_that("test ncol", { - # it works so how to test this? - - expect_equal(2L, 2L) -}) \ No newline at end of file diff --git a/tests/testthat/test-dplyr-verbs.r b/tests/testthat/test-dplyr-verbs.r deleted file mode 100644 index f8f18f35..00000000 --- a/tests/testthat/test-dplyr-verbs.r +++ /dev/null @@ -1,175 +0,0 @@ -context("test-dplyr-verbs") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_b_dv.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing select", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - df = b %>% - select(a) %>% - collect - - expect_equal(ncol(df), 1) -}) - -test_that("testing rename", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - df = b %>% - rename(a_new_name = a) %>% - collect - - expect_setequal(colnames(df), c("a_new_name", "b")) -}) - -test_that("testing filter", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - df = b %>% - filter(a <= 100, b <= 10) %>% - collect - - expect_setequal(nrow(df), 10) -}) - -test_that("testing filter - global vars", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - one_hundred = 100 - - df = b %>% - filter(a <= one_hundred, b <= 10) %>% - collect - - expect_setequal(nrow(df), 10) -}) - -test_that("testing mutate", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - - df = b %>% - mutate(d = a + b) %>% - collect - - expect_setequal(sum(df$d), sum(df$a, df$b)) - - df = b %>% - mutate(e = rank(desc(a))) %>% - collect - - expect_equal(nrow(df), 100) - - # need to test - value <- as.disk.frame(tibble(char = LETTERS, - num = 1:26)) - df2 = value %>% - dplyr::mutate(b = case_when( - char %in% c("A", "B", "C") ~ "1", - TRUE ~ char)) %>% - collect - - expect_equal(ncol(df2), 3) - - # testing - fn = function(a, b) { - a+b - } - - df3 = value %>% - dplyr::mutate(b = fn(num, num)) %>% - collect - - expect_equal(ncol(df3), 3) - - - global_var = 100 - - df4 = value %>% - dplyr::mutate(b = fn(num, num), d = global_var*2) %>% - collect - - expect_equal(ncol(df4), 4) - expect_true(all(df4$d == 200)) -}) - -test_that("testing mutate user-defined function", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - - udf = function(a1, b1) { - a1 + b1 - } - - df = b %>% - mutate(d = udf(a,b)) %>% - collect - - expect_setequal(sum(df$d), sum(df$a, df$b)) -}) - -test_that("testing transmute", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - df = b %>% - transmute(d = a + b) %>% - collect - - expect_setequal(names(df), c("d")) -}) - -test_that("testing arrange", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - expect_warning(df <- b %>% - mutate(random_unif = runif(dplyr::n())) %>% - arrange(desc(random_unif))) - - df <- b %>% - mutate(random_unif = runif(dplyr::n())) %>% - chunk_arrange(desc(random_unif)) - - x = purrr::map_lgl(1:nchunks(df), ~{ - is.unsorted(.x) == FALSE - }) - - expect_true(all(x)) -}) - -test_that("testing chunk_summarise", { - b = disk.frame(file.path(tempdir(), "tmp_b_dv.df")) - - df = b %>% - chunk_summarise(suma = sum(a)) %>% - collect %>% - summarise(suma = sum(suma)) - - expect_equal(df$suma, collect(b)$a %>% sum) -}) - -test_that("testing mutate within function works", { - test_f <- function(params, x_df){ - x_df %>% mutate(aha = params[1]*cyl + params[2]*disp) - } - - expect_true("aha" %in% names(test_f(c(1, 2), mtcars))) -}) - -test_that("filter failure: prevent github #191 regression", { - flights_df = as.disk.frame(nycflights13::flights) - - # expect error due to syntax error - expect_warning(expect_error(flights_df %>% - filter(tailnum %in% paste0(unique(nycflights13::flights$tailnum)[1:60]), "") %>% - collect)) - - delete(flights_df) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_b_dv.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-dtplyr-support.r b/tests/testthat/test-dtplyr-support.r deleted file mode 100644 index a3499435..00000000 --- a/tests/testthat/test-dtplyr-support.r +++ /dev/null @@ -1,46 +0,0 @@ -context("test-dtplyr-verbs") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - tf = file.path(tempdir(), "test-dtplyr.df") - as.disk.frame(b, outdir = tf, nchunks = 5, overwrite = TRUE) -}) - -test_that("testing dtplyr", { - # TODO add tests when new version of dtplyr on CRAN - # iris_df = as.disk.frame(iris) - # - # iris_df %>% - # filter(Sepal.Length > 7) %>% - # collect() - # - # - # aa = iris_df %>% - # cmap(~{ - # dtplyr::lazy_dt(.x) %>% - # filter(Sepal.Length > 7) %>% - # collect() - # }) %>% - # collect - # - # - # lazy_dt <- function(...) { - # UseMethod("lazy_dt") - # } - # - # lazy_dt.disk.frame <- function(df, ...) { - # cmap(df, ) - # } - # - # lazy_dt.default <- function(...) { - # dtplyr::lazy_dt(...) - # } - expect_true(TRUE) -}) - - - -teardown({ - fs::dir_delete(file.path(tempdir(), "test-dtplyr.df")) -}) - diff --git a/tests/testthat/test-foverlaps.r b/tests/testthat/test-foverlaps.r deleted file mode 100644 index a4438aa0..00000000 --- a/tests/testthat/test-foverlaps.r +++ /dev/null @@ -1,42 +0,0 @@ -context("test-foverlaps") - -setup({ - #setup_disk.frame(workers = 1) -}) - -# TODO currently it's not possible to do -test_that("test foverlap with data.frame", { - x = as.disk.frame(data.table(start=c(5,31,22,16), end=c(8,50,25,18), val2 = 7:10)) - y = as.disk.frame(data.table(start=c(10, 20, 30), end=c(15, 35, 45), val1 = 1:3)) - byxy = c("start", "end") - xy.df = foverlaps.disk.frame( - x, y, by.x = byxy, by.y = byxy, - merge_by_chunk_id = TRUE, overwrite = TRUE) - - collect(xy.df) - - testthat::expect_equal(nrow(xy.df), 3) -}) - - -# TODO this is also not a good test case -# test_that("test foverlap with disk.frame", { -# x = data.table(start=c(5,31,22,16), end=c(8,50,25,18), val2 = 7:10) -# y = data.table(start=c(10, 20, 30), end=c(15, 35, 45), val1 = 1:3) -# setkey(y, start, end) -# -# dx = shard(x, "tmp_fo.df", overwrite = T, shardby=c("start","end")) -# dy = shard(y, "tmp_to.df", overwrite = T, shardby=c("start","end")) -# -# xy1 = foverlaps(x,y, type="any", which = T) -# -# dxy1 = foverlaps.disk.frame(dx, dy, type="any", outdir="tmp_fo_out2.df") ## return overlap join -# dxy1c = dxy1 %>% collect -# -# foverlaps.disk.frame(dx, dy, type="any", mult="first", outdir="tmp_fo_out2.df") ## returns only first match -# foverlaps.disk.frame(dx, dy, type="within", outdir="tmp_fo_out3.df") ## matches iff 'x' is within 'y' -# }) - -teardown({ - -}) \ No newline at end of file diff --git a/tests/testthat/test-full_join.R b/tests/testthat/test-full_join.R deleted file mode 100644 index f40e8cb8..00000000 --- a/tests/testthat/test-full_join.R +++ /dev/null @@ -1,58 +0,0 @@ -context("test-full_join") - -setup({ - a = data.frame(a = 1:100, b = 1:100) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_fj.df"), nchunks = 4, overwrite = T) - as.disk.frame(b, file.path(tempdir(), "tmp_b_fj.df"), nchunks = 5, overwrite = T) - as.disk.frame(d, file.path(tempdir(), "tmp_d_fj.df"), overwrite = T) -}) - -test_that("testing full_join where right is data.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_fj.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_fj.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_fj.df")) - bc = collect(b) - dc = collect(d) - - abc <- full_join(a, bc, by = "a") %>% collect - expect_equal(nrow(abc), 150) - - abc0 <- full_join(a, bc, by = c("a","b")) %>% collect - expect_equal(nrow(abc0), 200) - - abc100 <- full_join(a, bc, by = "b") %>% collect - expect_equal(nrow(abc100), 100) - - abd50 <- full_join(a, dc, by = "b") %>% collect - expect_equal(nrow(abd50), 100) -}) - -test_that("testing full_join where right is disk.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_fj.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_fj.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_fj.df")) - - expect_warning({ - ab <- full_join(a, b, by = "a", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab), 150) - - expect_warning({ab0 = full_join(a, b, by = c("a","b"), merge_by_chunk_id = F) %>% collect}) - expect_equal(nrow(ab0), 200) - - expect_warning({ab100 = full_join(a, b, by = "b", merge_by_chunk_id = F) %>% collect}) - expect_equal(nrow(ab100), 100) - - expect_warning({ad50 = full_join(a, d, by = "b", merge_by_chunk_id = F) %>% collect}) - expect_equal(nrow(ad50), 100) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_a_fj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_fj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_fj.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-get_chunk.r b/tests/testthat/test-get_chunk.r deleted file mode 100644 index d33c336e..00000000 --- a/tests/testthat/test-get_chunk.r +++ /dev/null @@ -1,17 +0,0 @@ -context("test-get_chunk") - -setup({ - #setup_disk.frame(workers = 1) - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11), file.path(tempdir(), "tmp_del_delete"), overwrite=T) -}) - -test_that("data.table .N", { - df = disk.frame(file.path(tempdir(), "tmp_del_delete")) - expect_s3_class(get_chunk(df, 1), "data.frame") - - expect_s3_class(get_chunk(df, "1.fst"), "data.frame") -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_del_delete")) -}) \ No newline at end of file diff --git a/tests/testthat/test-get_chunk_ids.r b/tests/testthat/test-get_chunk_ids.r deleted file mode 100644 index 5e95c701..00000000 --- a/tests/testthat/test-get_chunk_ids.r +++ /dev/null @@ -1,20 +0,0 @@ -context("test-get_chunk_ids") - -setup({ - #setup_disk.frame(workers = 1) - df = as.disk.frame(disk.frame:::gen_datatable_synthetic(1e5+11), file.path(tempdir(), "tmp_del_delete"), overwrite=T) -}) - -test_that("get_chunk_ids", { - df = disk.frame(file.path(tempdir(), "tmp_del_delete")) - - gci = get_chunk_ids(df) - expect_type(get_chunk_ids(df), "character") - - gcis = get_chunk_ids(df, strip_extension = F) - expect_true("1.fst" %in% gcis) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_del_delete")) -}) \ No newline at end of file diff --git a/tests/testthat/test-glm.r b/tests/testthat/test-glm.r deleted file mode 100644 index efa0122a..00000000 --- a/tests/testthat/test-glm.r +++ /dev/null @@ -1,30 +0,0 @@ -context("test-glm") - -setup({ - #setup_disk.frame(workers = 1) -}) - -test_that("glm", { - cars.df = as.disk.frame(cars, outdir = file.path(tempdir(), "cars.df"), overwrite = TRUE) - - majorv = as.integer(version$major) - minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1]) - - if((majorv == 3) & (minorv < 6)) { - expect_warning({m <- dfglm(dist~speed, cars.df, glm_backend = "biglm")}) - } else { - m <- dfglm(dist~speed, cars.df, glm_backend = "biglm") - } - summary(m) - - if((majorv == 3) & (minorv >= 6) ) { - broom::tidy(m) - } - - m <- dfglm(dist~speed, cars.df, glm_backend = "speedglm") - summary(m) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "cars.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-group-by.R b/tests/testthat/test-group-by.R deleted file mode 100644 index d6684114..00000000 --- a/tests/testthat/test-group-by.R +++ /dev/null @@ -1,309 +0,0 @@ -context("test-group_by") - -setup({ - df = disk.frame:::gen_datatable_synthetic(1e3+11) - data.table::fwrite(df, file.path(tempdir(), "tmp_pls_delete_gb.csv")) -}) - - -test_that("new group_by framework", { - if(interactive()) { - iris.df = iris %>% - as.disk.frame - - grpby = iris.df %>% - group_by(Species) %>% - summarize(mean(Petal.Length), sumx = sum(Petal.Length/Sepal.Width), sd(Sepal.Width/ Petal.Length), var(Sepal.Width/ Sepal.Width)) %>% - collect - - grpby2 = iris %>% - group_by(Species) %>% - summarize(mean(Petal.Length), sumx = sum(Petal.Length/Sepal.Width), sd(Sepal.Width/ Petal.Length), var(Sepal.Width/ Sepal.Width)) %>% - arrange() - - for (n in names(grpby)) { - expect_true(all(grpby2[, n] == grpby[, n]) || all(abs(grpby2[, n] - grpby[, n]) < 0.0001)) - } - - delete(iris.df) - } - expect_true(TRUE) -}) - -test_that("new group_by framework - no group-by just summarise", { - if(interactive()) { - iris.df = iris %>% - as.disk.frame - - grpby = iris.df %>% - summarize(mean(Petal.Length), sumx = sum(Petal.Length/Sepal.Width), sd(Sepal.Width/ Petal.Length), var(Sepal.Width/ Sepal.Width)) %>% - collect - - grpby2 = iris %>% - summarize(mean(Petal.Length), sumx = sum(Petal.Length/Sepal.Width), sd(Sepal.Width/ Petal.Length), var(Sepal.Width/ Sepal.Width)) %>% - arrange() - - for (n in names(grpby)) { - expect_true(all(grpby2[, n] == grpby[, n]) || all(abs(grpby2[, n] - grpby[, n]) < 0.0001)) - } - - delete(iris.df) - } - expect_true(TRUE) -}) - -# test_that("new group_by framework - nested-group-by", { - # if(interactive()) { - # iris.df = iris %>% - # as.disk.frame - # - # expect_error(grpby <- iris.df %>% - # summarize(mean(Petal.Length + max(Petal.Length))) %>% - # collect) - # - # expect_error(grpby <- iris.df %>% - # summarize(mean(Petal.Length) + max(Petal.Length)) %>% - # collect) - # - # expect_error(grpby <- iris.df %>% - # summarize(mean(Petal.Length) + 1) %>% - # collect) - # - # expect_error(grpby <- iris.df %>% - # summarize(list(mean(Petal.Length))) %>% - # collect) - # - # fn_tmp = function(x) x + 1 - # grpby <- iris.df %>% - # summarize(mean(fn_tmp(Petal.Length))) %>% - # collect - # - # grpby2 <- iris %>% - # summarize(mean(fn_tmp(Petal.Length))) - # - # for (n in names(grpby)) { - # expect_true(all(grpby2[, n] == grpby[, n]) || all(abs(grpby2[, n] - grpby[, n]) < 0.0001)) - # } - # delete(iris.df) - # } - # expect_true(TRUE) -# }) - -test_that("guard against github #241", { - if(interactive()) { - # I suspect there was an issue with number of chunk = 1 - result_from_disk.frame = iris %>% - as.disk.frame(nchunks = 1) %>% - group_by(Species) %>% - summarize( - mean(Petal.Length), - sumx = sum(Petal.Length/Sepal.Width), - sd(Sepal.Width/ Petal.Length), - var(Sepal.Width/ Sepal.Width), - l = length(Sepal.Width/ Sepal.Width + 2), - max(Sepal.Width), - min(Sepal.Width), - median(Sepal.Width) - ) %>% - collect - } else { - expect_true(TRUE) - } -}) - - -test_that("group_by", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - - dff_res = dff %>% - collect %>% - group_by(id1) %>% - summarise(mv1 = mean(v1)) - - dff1 <- dff %>% - chunk_group_by(id1, id2) %>% - chunk_summarise(mv1 = mean(v1)) %>% - collect - - expect_false(nrow(dff1) == nrow(dff_res)) -}) - -test_that("test hard_group_by on disk.frame", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - - dff_res = dff %>% - collect %>% - group_by(id1, id2) %>% - summarise(mv1 = mean(v1)) - - dff1 <- dff %>% - hard_group_by(id1, id2) %>% - chunk_summarise(mv1 = mean(v1)) %>% collect - - expect_equal(nrow(dff1), nrow(dff_res)) -}) - -test_that("test hard_group_by on data.frame", { - df = disk.frame:::gen_datatable_synthetic(1e3+11) - - df1 = df %>% - group_by(id1, id2) %>% - summarise(mv1 = mean(v1)) - - dff1 <- df %>% - hard_group_by(id1,id2) %>% - summarise(mv1 = mean(v1)) - - expect_equal(nrow(dff1), nrow(df1)) -}) - - -test_that("test hard_group_by on disk.frame (sort)", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - - dff_res = dff %>% - collect %>% - group_by(id1, id2) %>% - summarise(mv1 = mean(v1)) - - dff1 <- dff %>% - hard_group_by(id1, id2, shardby_function="sort") %>% - chunk_summarise(mv1 = mean(v1)) %>% collect - - expect_equal(nrow(dff1), nrow(dff_res)) -}) - -test_that("test hard_group_by on data.frame (sort)", { - df = disk.frame:::gen_datatable_synthetic(1e3+11) - - df1 = df %>% - group_by(id1, id2) %>% - summarise(mv1 = mean(v1)) - - dff1 <- df %>% - hard_group_by(id1, id2, shardby_function="sort") %>% - summarise(mv1 = mean(v1)) - - expect_equal(nrow(dff1), nrow(df1)) -}) - -test_that("guard against github 256", { - test2 <- tibble::tibble( - date = lubridate::ymd(rep(c("2019-01-02", "2019-02-03", "2019-03-04"), 4)), - uid = as.factor(rep(c(uuid::UUIDgenerate(), uuid::UUIDgenerate()), 6)), - proto = as.factor(rep(c("TCP", "UDP", "ICMP"), 4)), - port = as.double(rep(c(22, 21, 0), 4)) - ) - - correct_result = test2 %>% - group_by(date, uid, proto, port) %>% - summarize(n=n()) %>% - collect - - test_df = as.disk.frame(test2, nchunks = 2, overwrite=TRUE) - - incorrect_result = test_df %>% - group_by(date, uid, proto, port) %>% - summarize(n=n()) %>% - collect - - expect_equal(dim(incorrect_result), dim(correct_result)) -}) - -test_that("guard against github 256 #2", { - test2 <- tibble::tibble( - date = lubridate::ymd(rep(c("2019-01-02", "2019-02-03", "2019-03-04"), 4)), - uid = as.factor(rep(c(uuid::UUIDgenerate(), uuid::UUIDgenerate()), 6)), - proto = as.factor(rep(c("TCP", "UDP", "ICMP"), 4)), - port = as.double(rep(c(22, 21, 0), 4)) - ) - - test_df = as.disk.frame(test2, nchunks = 2, overwrite=TRUE) - - - correct_result = test_df %>% - group_by(!!!syms(names(test_df))) %>% - summarize(n=n()) %>% - collect - - incorrect_result = test_df %>% - group_by(date, uid, proto, port) %>% - summarize(n=n()) %>% - collect - - expect_equal(dim(incorrect_result), dim(correct_result)) -}) - -test_that("guard against github 256 #3", { - library(testthat) - library(disk.frame) - setup_disk.frame() - - test2 <- tibble::tibble( - date = sample(1:10, 20, replace = TRUE), - uid = sample(1:10, 20, replace = TRUE) - ) - - test_df = as.disk.frame(test2, nchunks = 2, overwrite=TRUE) - - ntd = names(test_df) - - correct_result = test_df %>% - group_by(!!!syms(ntd)) %>% - summarize(n=n()) %>% - collect - - incorrect_result = test_df %>% - group_by(date, uid) %>% - summarize(n=n()) %>% - collect - - expect_equal(dim(incorrect_result), dim(correct_result)) -}) - -test_that("tests for github #250", { - aggregate_expressions <- list(n = quote(n())) - - result1 = iris %>% - as.disk.frame() %>% - group_by(Species) %>% - summarise(n = n()) %>% - collect - - result2 <- iris %>% - as.disk.frame() %>% - group_by(Species) %>% - summarize(!!!(aggregate_expressions)) %>% - collect - - expect_equal(result1, result2) -}) - -test_that("tests for github #250 2", { - aggregate_expressions <- list(n = quote(n()), quote(n())) - - result1 = iris %>% - as.disk.frame() %>% - group_by(Species) %>% - summarise(n = n(), n()) %>% - collect; result1 - - result2 <- iris %>% - as.disk.frame() %>% - group_by(Species) %>% - summarize(!!!(aggregate_expressions)) %>% - collect - - expect_equal(result1, result2) -}) - -teardown({ - fs::file_delete(file.path(tempdir(), "tmp_pls_delete_gb.csv")) - fs::dir_delete(file.path(tempdir(), "tmp_pls_delete_gb.df")) -}) diff --git a/tests/testthat/test-hard-arrange.R b/tests/testthat/test-hard-arrange.R deleted file mode 100644 index 0419cee8..00000000 --- a/tests/testthat/test-hard-arrange.R +++ /dev/null @@ -1,111 +0,0 @@ -context("test-arrange") - -setup({ - - df = disk.frame:::gen_datatable_synthetic(1e3+11) - data.table::fwrite(df, file.path(tempdir(), "tmp_pls_delete_gb.csv")) -}) - -test_that("test hard_arrange on disk.frame, single chunk", { - # Randomise rows since rows are already sorted - iris.df = as.disk.frame(sample_n(iris, nrow(iris)), nchunks = 1) - iris_hard.df = hard_arrange(iris.df, Species) - - # Check sort - expect_true(!is.unsorted(iris_hard.df$Species)) -}) - -test_that("test hard_arrange on disk.frame, single variable", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - - # Sort ascending, one level - sorted_dff <- dff %>% hard_arrange(id1) - sorted_df <- sorted_dff %>% collect - - expect_true(!is.unsorted(sorted_df$id1)) -}) - -test_that("test hard_arrange on disk.frame, factor data type", { - iris.df = as.disk.frame(sample_n(iris, nrow(iris)), nchunks = 2) - iris_hard.df = hard_arrange(iris.df, Species) - - expect_true(!is.unsorted(iris_hard.df$Species)) -}) - -test_that("test hard_arrange on disk.frame, date data type", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - sorted_dff <- dff %>% hard_arrange(date1) - - expect_true(!is.unsorted(sorted_dff$date1)) -}) - -test_that("test hard_arrange on disk.frame, two and three variables", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - - dfp <- read.csv(file.path(tempdir(), "tmp_pls_delete_gb.csv")) - - # Sort ascending, two levels - sorted_dff <- dff %>% hard_arrange(id1, id4) %>% collect - sorted_dfp <- dff %>% collect %>% dplyr::arrange(id1, id4) - - # Compare vs dplyr - expect_true(all(sorted_dff$id1 == sorted_dfp$id1)) - expect_true(all(sorted_dff$id4 == sorted_dfp$id4)) - - # Sort ascending, three levels, from already partially sorted disk frame - sorted_dff2 <- sorted_dff %>% hard_arrange(id1, id4, id6) %>% collect - sorted_dfp2 <- dff %>% collect %>% dplyr::arrange(id1, id4, id6) - - # Compare vs dplyr - expect_true(all(sorted_dff2$id1 == sorted_dfp2$id1)) - expect_true(all(sorted_dff2$id4 == sorted_dfp2$id4)) - expect_true(all(sorted_dff2$id6 == sorted_dfp2$id6)) -}) - -test_that("test hard_arrange on disk.frame, two factors", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete_gb.csv"), - file.path(tempdir(), "tmp_pls_delete_gb.df")) - - # Sort decending, two levels - desc_dff <- dff %>% hard_arrange(desc(id4), id2) - desc_dff <- desc_dff %>% collect - - # Level 1 - expect_true(!is.unsorted(-desc_dff$id4)) - - # Level 2 - desc_dff$id4_id2 <- paste0( - formatC(max(desc_dff$id4) - desc_dff$id4, width=3, format="d", flag= "0"), - desc_dff$id2) - expect_true(!is.unsorted(-desc_dff$id4)) -}) - -test_that("test hard_arrange on data.frame vs dplyr", { - df = disk.frame:::gen_datatable_synthetic(1e3+11) - - # Sort ascending - sorted_dff <- df %>% hard_arrange(id1, id4) %>% collect - sorted_dfp <- df %>% dplyr::arrange(id1, id4) - - expect_true(all(sorted_dff$id1 == sorted_dfp$id1)) - expect_true(all(sorted_dff$id4 == sorted_dfp$id4)) - - # Sort decending - desc_dff <- df %>% hard_arrange(desc(id4), id2) %>% collect - desc_dfp <- df %>% dplyr::arrange(desc(id4), id2) - - expect_true(all(sorted_dff$id4 == sorted_dfp$id4)) - expect_true(all(sorted_dff$id2 == sorted_dfp$dfp)) -}) - -teardown({ - fs::file_delete(file.path(tempdir(), "tmp_pls_delete_gb.csv")) - fs::dir_delete(file.path(tempdir(), "tmp_pls_delete_gb.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-inner_join.R b/tests/testthat/test-inner_join.R deleted file mode 100644 index e165a142..00000000 --- a/tests/testthat/test-inner_join.R +++ /dev/null @@ -1,64 +0,0 @@ -context("test-inner_join") - -setup({ - a = data.frame(a = 1:100, b = 1:100) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_ij.df"), nchunks = 4, overwrite = T) - as.disk.frame(b, file.path(tempdir(), "tmp_b_ij.df"), nchunks = 5, overwrite = T) - as.disk.frame(d, file.path(tempdir(), "tmp_d_ij.df"), overwrite = T) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_ij2.df"), nchunks = 4, overwrite = T) - as.disk.frame(b, file.path(tempdir(), "tmp_b_ij2.df"), nchunks = 5, overwrite = T) - as.disk.frame(d, file.path(tempdir(), "tmp_d_ij2.df"), overwrite = T) -}) - -test_that("testing inner_join where right is data.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_ij.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_ij.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_ij.df")) - bc = collect(b) - dc = collect(d) - - abc = inner_join(a, bc, by = "a") %>% collect - expect_equal(nrow(abc), 50) - - abc0 = inner_join(a, bc, by = c("a","b")) %>% collect - expect_equal(nrow(abc0), 0) - - abc100 = inner_join(a, bc, by = "b") %>% collect - expect_equal(nrow(abc100), 100) - - abd50 = inner_join(a, dc, by = "b") %>% collect - expect_equal(nrow(abd50), 50) -}) - -test_that("testing inner_join where right is disk.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_ij2.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_ij2.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_ij2.df")) - - ab = inner_join(a, b, by = "a", merge_by_chunk_id = F) %>% collect - expect_equal(nrow(ab), 50) - - # expecting a warning for some chunks being 0 rows - expect_warning(ab0 <- inner_join(a, b, by = c("a","b"), merge_by_chunk_id = F) %>% collect) - expect_equal(nrow(ab0), 0) - - ab100 = inner_join(a, b, by = "b", merge_by_chunk_id = F) %>% collect - expect_equal(nrow(ab100), 100) - - ad50 = inner_join(a, d, by = "b", merge_by_chunk_id = F) %>% collect - expect_equal(nrow(ad50), 50) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_a_ij.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_ij.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_ij.df")) - fs::dir_delete(file.path(tempdir(), "tmp_a_ij2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_ij2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_ij2.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-is-disk.frame.r b/tests/testthat/test-is-disk.frame.r deleted file mode 100644 index 87d4fde4..00000000 --- a/tests/testthat/test-is-disk.frame.r +++ /dev/null @@ -1,12 +0,0 @@ -context("test-is-disk-frame") - -test_that("testing is_disk.frame", { - fs::dir_create(file.path(tempdir(), "tmp_is_disk_frame")) - fst::write_fst(data.frame(a= 1, b = 1), file.path(tempdir(), "tmp_is_disk_frame/1.fst")) - fst::write_fst(data.frame(a= 1, b = 1), file.path(tempdir(), "tmp_is_disk_frame/2.fst")) - - df = disk.frame(file.path(tempdir(), "tmp_is_disk_frame")) - expect_true(is_disk.frame(df)) - - disk.frame::delete(df) -}) diff --git a/tests/testthat/test-left_join.R b/tests/testthat/test-left_join.R deleted file mode 100644 index 66801085..00000000 --- a/tests/testthat/test-left_join.R +++ /dev/null @@ -1,71 +0,0 @@ -context("test-left_join") - -setup({ - - a = data.frame(a = 1:100, b = 1:100) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_lj.df"), nchunks = 4, overwrite = T) - as.disk.frame(b, file.path(tempdir(), "tmp_b_lj.df"), nchunks = 5, overwrite = T) - as.disk.frame(d, file.path(tempdir(), "tmp_d_lj.df"), overwrite = T) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_lj2.df"), nchunks = 4, overwrite = T) - as.disk.frame(b, file.path(tempdir(), "tmp_b_lj2.df"), nchunks = 5, overwrite = T) - as.disk.frame(d, file.path(tempdir(), "tmp_d_lj2.df"), overwrite = T) -}) - -test_that("testing left_join where right is data.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_lj.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_lj.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_lj.df")) - bc = collect(b) - dc = collect(d) - - abc = left_join(a, bc, by = "a") %>% collect - expect_equal(nrow(abc), 100) - - abc0 = left_join(a, bc, by = c("a","b")) %>% collect - expect_equal(nrow(abc0), 100) - - abc100 = left_join(a, bc, by = "b") %>% collect - expect_equal(nrow(abc100), 100) - - abd50 = left_join(a, dc, by = "b") %>% collect - expect_equal(nrow(abd50), 100) -}) - -test_that("testing left_join where right is disk.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_lj2.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_lj2.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_lj2.df")) - - expect_warning({ - ab = left_join(a, b, by = "a", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab), 100) - - expect_warning({ - ab0 = left_join(a, b, by = c("a","b"), merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab0), 100) - - expect_warning({ - ab100 = left_join(a, b, by = "b", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab100), 100) - - expect_warning({ - ad50 = left_join(a, d, by = "b", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ad50), 100) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_a_lj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_lj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_lj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_a_lj2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_lj2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_lj2.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-map.r b/tests/testthat/test-map.r deleted file mode 100644 index beec44ba..00000000 --- a/tests/testthat/test-map.r +++ /dev/null @@ -1,76 +0,0 @@ -context("test-cmap") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_map.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing cmap lazy", { - b = disk.frame(file.path(tempdir(), "tmp_map.df")) - - # return 1 row from each chunk - df = b %>% cmap(~.x[1]) - - expect_s3_class(df, "disk.frame") - - df2 = df %>% collect - - expect_s3_class(df2, "data.frame") - - expect_equal(nrow(df2), 5L) -}) - -test_that("testing cmap eager", { - b = disk.frame(file.path(tempdir(), "tmp_map.df")) - - # return 1 row from each chunk - df = b %>% cmap(~.x[1], lazy = F) - expect_false("disk.frame" %in% class(df)) - - # return 1 row from each chunk - df = b %>% cmap_dfr(~.x[1]) - expect_false("disk.frame" %in% class(df)) - expect_true("data.frame" %in% class(df)) -}) - -test_that("testing delayed", { - b = disk.frame(file.path(tempdir(), "tmp_map.df")) - - # return 1 row from each chunk - df = b %>% delayed(~.x[1]) - - expect_s3_class(df, "disk.frame") - - df1 = collect(df) - - expect_equal(nrow(df1), 5) -}) - - -test_that("testing map_dfr", { - b = disk.frame(file.path(tempdir(), "tmp_map.df")) - - # return 1 row from each chunk - df = b %>% cmap_dfr(~.x[1,]) - - expect_s3_class(df, "data.frame") -}) - - -test_that("testing imap", { - b = disk.frame(file.path(tempdir(), "tmp_map.df")) - - # return 1 row from each chunk - df = b %>% cimap_dfr(~{ - y = .x[1,] - y[,ok := .y] - y - }) - - expect_s3_class(df, "data.frame") -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_map.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-map2.r b/tests/testthat/test-map2.r deleted file mode 100644 index 926d49ee..00000000 --- a/tests/testthat/test-map2.r +++ /dev/null @@ -1,46 +0,0 @@ -context("test-cmap2") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 151:250, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_map2.df"), nchunks = 5, overwrite = T) - as.disk.frame(d, file.path(tempdir(), "tmp_map2d.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing cmap2 .y is disk.frame", { - b = disk.frame(file.path(tempdir(), "tmp_map2.df")) - d = disk.frame(file.path(tempdir(), "tmp_map2d.df")) - - # return 1 row from each chunk - df = cmap2(b, d, ~rbindlist(list(.x[1,],.y[1,])), outdir = file.path(tempdir(), "tmp_map2_out.df")) - - expect_s3_class(df, "disk.frame") - - df2 = df %>% collect - - expect_s3_class(df2, "data.frame") - - expect_equal(nrow(df2), 10L) -}) - -test_that("testing map2 .y is not disk.frame", { - b = disk.frame(file.path(tempdir(), "tmp_map2.df")) - d = 1:nchunks(b) - - # return 1 row from each chunk - expect_warning(df <- cmap2(b, d, ~.x[1,.(y = .y)], outdir = "tmp_map2_out2.df")) - - expect_type(df, "list") - - df2 = df %>% rbindlist - - expect_s3_class(df2, "data.frame") - - expect_equal(nrow(df2), 5L) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_map2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_map2d.df")) - fs::dir_delete(file.path(tempdir(), "tmp_map2_out.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-merge.r b/tests/testthat/test-merge.r deleted file mode 100644 index 41bb37e4..00000000 --- a/tests/testthat/test-merge.r +++ /dev/null @@ -1,59 +0,0 @@ -context("test-merge.disk.frame") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 151:250, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_merge.df"), nchunks = 5, overwrite = TRUE) - as.disk.frame(d, file.path(tempdir(), "tmp_merge2.df"), nchunks = 5, overwrite = TRUE) -}) - -test_that("testing merge of disk.frame", { - b.df = disk.frame(file.path(tempdir(), "tmp_merge.df")) - d.df = disk.frame(file.path(tempdir(), "tmp_merge2.df")) - - bd.df = merge(b.df, d.df, by = "b", outdir = file.path(tempdir(), "tmp_bd_merge.df"), overwrite = TRUE, merge_by_chunk_id = TRUE) - - expect_s3_class(bd.df, "disk.frame") - expect_equal(nrow(bd.df), 100) -}) - -test_that("testing merge of data.frame", { - b.df = disk.frame(file.path(tempdir(), "tmp_merge.df")) - d = data.frame(a = 151:250, b = 1:100) - - bd.df = merge(b.df, d, by = "b", outdir = file.path(tempdir(), "tmp_bd_merge2.df"), overwrite = TRUE) - - expect_s3_class(bd.df, "disk.frame") - expect_equal(nrow(bd.df), 100) - - tmp = collect(bd.df) - - expect_s3_class(tmp, "data.frame") - expect_equal(nrow(tmp), 100) -}) - -test_that("testing error when merge_by_chunk = FALSE", { - b.df = disk.frame(file.path(tempdir(), "tmp_merge.df")) - d.df = disk.frame(file.path(tempdir(), "tmp_merge2.df")) - - testthat::expect_error() - expect_error( - merge( - b.df, - d.df, - by = "b", - outdir = file.path(tempdir(), "tmp_bd_merge.df"), - overwrite = TRUE, - merge_by_chunkd_id = FALSE - ) - ) -}) - - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_merge.df")) - fs::dir_delete(file.path(tempdir(), "tmp_merge2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_bd_merge.df")) - fs::dir_delete(file.path(tempdir(), "tmp_bd_merge2.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-names.r b/tests/testthat/test-names.r deleted file mode 100644 index 2d114aeb..00000000 --- a/tests/testthat/test-names.r +++ /dev/null @@ -1,25 +0,0 @@ -context("test-names") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_names.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing names", { - b = disk.frame(file.path(tempdir(), "tmp_names.df")) - - expect_setequal(colnames(b), c("a","b")) - expect_setequal(names(b), c("a","b")) -}) - -test_that("testing names with lazyfn", { - b = disk.frame(file.path(tempdir(), "tmp_names.df")) %>% - mutate(d = a + b) - - expect_setequal(colnames(b), c("a","b", "d")) - expect_setequal(names(b), c("a","b", "d")) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_names.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-nchunks.r b/tests/testthat/test-nchunks.r deleted file mode 100644 index bf716810..00000000 --- a/tests/testthat/test-nchunks.r +++ /dev/null @@ -1,18 +0,0 @@ -context("test-nchunks") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_chunks.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing nchunks", { - b = disk.frame(file.path(tempdir(), "tmp_chunks.df")) - - expect_equal(nchunks(b), 5) - expect_equal(nchunk(b), 5) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_chunks.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-nrow-ncol.R b/tests/testthat/test-nrow-ncol.R deleted file mode 100644 index 05b33601..00000000 --- a/tests/testthat/test-nrow-ncol.R +++ /dev/null @@ -1,20 +0,0 @@ -context("test-nrow-ncol") - -setup({ - df = disk.frame:::gen_datatable_synthetic(1e3+11) - data.table::fwrite(df, file.path(tempdir(), "tmp_pls_delete.csv")) -}) - -test_that("nrow ncol", { - dff = csv_to_disk.frame( - file.path(tempdir(), "tmp_pls_delete.csv"), - file.path(tempdir(), "tmp_pls_delete.df")) - - expect_equal(nrow(dff), 1e3+11) - expect_equal(ncol(dff), 10) -}) - -teardown({ - fs::file_delete(file.path(tempdir(), "tmp_pls_delete.csv")) - fs::dir_delete(file.path(tempdir(), "tmp_pls_delete.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-overwrite_check.r b/tests/testthat/test-overwrite_check.r deleted file mode 100644 index 59f6548e..00000000 --- a/tests/testthat/test-overwrite_check.r +++ /dev/null @@ -1,21 +0,0 @@ -context("test-overwrite_check") - -setup({ -}) - -test_that("testing overwrite_check", { - b = data.frame(a = 51:150, b = 1:100) - - fs::dir_create(file.path(tempdir(), "tmp_overwrite-check")) - fs::file_create(file.path(tempdir(), "tmp_overwrite-check/tmp")) - - - expect_error(disk.frame::overwrite_check(file.path(tempdir(), "tmp_overwrite-check"), overwrite = TRUE)) - - expect_error(disk.frame::overwrite_check(file.path(tempdir(), "tmp_overwrite-check"), overwrite = FALSE)) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_overwrite-check")) -}) \ No newline at end of file diff --git a/tests/testthat/test-pls-add.r b/tests/testthat/test-pls-add.r deleted file mode 100644 index 29f0e78b..00000000 --- a/tests/testthat/test-pls-add.r +++ /dev/null @@ -1,44 +0,0 @@ -context("test-pls-add") - -setup({ - #setup_disk.frame(workers = 1) -}) - -test_that("pls-add", { - - if (interactive()) { - library(disk.frame) - library(tidyverse) - - setup_disk.frame(2) - example <- as.disk.frame( - data.frame( - purchase_date=c("2020-03-20","2020-04-20"), - a = 1:2, - b = 3:4 - ) - ) - example %>% - mutate(Panel_Month = str_sub(purchase_date, 6, 7)) %>% - collect - - str_sub2 = function(xx, yy) xx + yy - - example %>% - mutate(Panel_Month = str_sub2(a, b)) %>% - collect - - example %>% - mutate(Panel_Month = str_sub2(a, 7)) %>% - collect - - example %>% - mutate(Panel_Month = str_sub2(6, 7)) %>% - collect - - - example %>% - mutate(Panel_Month = str_subs(purchase_date, 7)) %>% - collect - } -}) \ No newline at end of file diff --git a/tests/testthat/test-print.r b/tests/testthat/test-print.r deleted file mode 100644 index 9d1053cc..00000000 --- a/tests/testthat/test-print.r +++ /dev/null @@ -1,14 +0,0 @@ -context("test-print") - -setup({ -}) - -test_that("testing print", { - # TODO proper tests - expect_equal(2L, 2L) -}) - - -teardown({ - -}) \ No newline at end of file diff --git a/tests/testthat/test-pull.r b/tests/testthat/test-pull.r deleted file mode 100644 index 4acfa8d4..00000000 --- a/tests/testthat/test-pull.r +++ /dev/null @@ -1,50 +0,0 @@ -context("test-pull") - - -test_that("pull with", { - flights_df = as.disk.frame(nycflights13::flights) - - a = flights_df %>% - pull(carrier, carrier) - - b = flights_df %>% collect() %>% pull(carrier, carrier) - - expect_equal(a, b) - - a = flights_df %>% - pull(2, 2) - b = flights_df %>% collect() %>% pull(2, 2) - - expect_equal(a, b) - - a = flights_df %>% - pull(-1, -1) - b = flights_df %>% collect() %>% pull(-1, -1) - expect_equal(a, b) - - delete(flights_df) -}) - - -test_that("pull", { - flights_df = as.disk.frame(nycflights13::flights) - - a = flights_df %>% - pull(carrier) - b = flights_df %>% collect() %>% pull(carrier) - - expect_setequal(a, b) - - a = flights_df %>% - pull(2) - b = flights_df %>% collect() %>% pull(2) - - expect_setequal(a, b) - - a = flights_df %>% - pull(-1) - b = flights_df %>% collect() %>% pull(-1) - expect_setequal(a, b) - - delete(flights_df) -}) diff --git a/tests/testthat/test-rbindlist.r b/tests/testthat/test-rbindlist.r deleted file mode 100644 index d30cada0..00000000 --- a/tests/testthat/test-rbindlist.r +++ /dev/null @@ -1,31 +0,0 @@ -context("test-rbindlist") - -setup({ - as.disk.frame(disk.frame:::gen_datatable_synthetic(1e3+11), file.path(tempdir(), "tmp_rbindlist1.df"), overwrite=TRUE) - as.disk.frame(disk.frame:::gen_datatable_synthetic(1e3+11), file.path(tempdir(), "tmp_rbindlist2.df"), overwrite=TRUE) - as.disk.frame(disk.frame:::gen_datatable_synthetic(1e3+11), file.path(tempdir(), "tmp_rbindlist4.df"), overwrite=TRUE) -}) - -test_that("test rbindlist", { - df1 = disk.frame(file.path(tempdir(), "tmp_rbindlist1.df")) - df2 = disk.frame(file.path(tempdir(), "tmp_rbindlist2.df")) - - df3 = rbindlist.disk.frame(list(df1, df2), outdir = file.path(tempdir(), "tmp_rbindlist3.df"), overwrite=TRUE) - - expect_equal(nrow(df3), 2*(1e3+11)) -}) - -test_that("test rbindlist accepts only list", { - df1 = disk.frame(file.path(tempdir(), "tmp_rbindlist4.df")) - - expect_error(rbindlist.disk.frame(df1, outdir = file.path(tempdir(), "tmp_rbindlist5.df"))) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_rbindlist1.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rbindlist2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rbindlist3.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rbindlist4.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rbindlist5.df")) -}) diff --git a/tests/testthat/test-rechunk.r b/tests/testthat/test-rechunk.r deleted file mode 100644 index 68f59938..00000000 --- a/tests/testthat/test-rechunk.r +++ /dev/null @@ -1,82 +0,0 @@ -context("test-rechunk") - -setup({ -}) - -test_that("testing rechunk 5 to 4", { - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_rechunks.df"), nchunks = 5, overwrite = T) - - b = disk.frame(file.path(tempdir(), "tmp_rechunks.df")) - - b = rechunk(b, 4) - expect_equal(nrow(b), 100) - expect_equal(ncol(b), 2) - expect_equal(nchunk(b), 4) - - res = collect(b)[order(b)] - - expect_equal(res$b, 1:100) - expect_equal(res$a, 51:150) -}) - -test_that("testing rechunk 5 to 3", { - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_rechunks2.df"), nchunks = 5, overwrite = T) - - b = disk.frame(file.path(tempdir(), "tmp_rechunks2.df")) - - b = rechunk(b, 3) - expect_equal(nrow(b), 100) - expect_equal(ncol(b), 2) - expect_equal(nchunk(b), 3) - - res = collect(b)[order(b)] - - expect_equal(res$b, 1:100) - expect_equal(res$a, 51:150) -}) - -test_that("testing rechunk 5 to 6", { - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_rechunks3.df"), nchunks = 5, overwrite = T) - - b = disk.frame(file.path(tempdir(), "tmp_rechunks3.df")) - - b = rechunk(b, 6) - expect_equal(nrow(b), 100) - expect_equal(ncol(b), 2) - expect_equal(nchunk(b), 6) - - res = collect(b)[order(b)] - - expect_equal(res$b, 1:100) - expect_equal(res$a, 51:150) -}) - -test_that("testing rechunk 5 to 7", { - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_rechunks4.df"), nchunks = 5, overwrite = T) - - b = disk.frame(file.path(tempdir(), "tmp_rechunks4.df")) - - b = rechunk(b, 7) - expect_equal(nrow(b), 100) - expect_equal(ncol(b), 2) - expect_equal(nchunk(b), 7) - - res = collect(b)[order(b)] - - expect_equal(res$b, 1:100) - expect_equal(res$a, 51:150) -}) - -# TODO do shardby; it's kinda of mitigated by thorough testing on Fannie Mae - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_rechunks.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rechunks2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rechunks3.df")) - fs::dir_delete(file.path(tempdir(), "tmp_rechunks4.df")) -}) diff --git a/tests/testthat/test-recommend_nchunk.R b/tests/testthat/test-recommend_nchunk.R deleted file mode 100644 index c5fbb109..00000000 --- a/tests/testthat/test-recommend_nchunk.R +++ /dev/null @@ -1,16 +0,0 @@ -context("test-recommend_nchunk") - -test_that("testing df_ram_size", { - expect_true(is.numeric(df_ram_size())) - - expect_true(!is.na(df_ram_size())) - expect_true(!is.null(df_ram_size())) - expect_true(!is.nan(df_ram_size())) - expect_true(is.finite(df_ram_size())) -}) - -test_that("testing df_ram_size; guards #213", { - # TODO tests - expect_true(df_ram_size() >= 1) -}) - diff --git a/tests/testthat/test-remove_chunk.r b/tests/testthat/test-remove_chunk.r deleted file mode 100644 index 8aec02ae..00000000 --- a/tests/testthat/test-remove_chunk.r +++ /dev/null @@ -1,18 +0,0 @@ -context("test-remove") - -test_that("testing remove chunk 3 of 5", { - b = data.frame(a = 51:150, b = 1:100) - tmp = file.path(tempdir(), "tmp_remove.df") - b = as.disk.frame(b, tmp, nchunks = 5, overwrite = T) - - b = remove_chunk(b, 3) - expect_equal(nrow(b), 80) - expect_equal(ncol(b), 2) - expect_equal(nchunk(b), 4) - - res <- collect(b)[order(b)] - - expect_equal(nrow(res), 80) - - delete(b) -}) diff --git a/tests/testthat/test-right_join.r b/tests/testthat/test-right_join.r deleted file mode 100644 index 27e532b4..00000000 --- a/tests/testthat/test-right_join.r +++ /dev/null @@ -1,7 +0,0 @@ -context("test-right_join") - -test_that("testing right_join", { - # TODO tests - expect_equal(2L, 2L) -}) - diff --git a/tests/testthat/test-sample_frac.r b/tests/testthat/test-sample_frac.r deleted file mode 100644 index 8c45f8f7..00000000 --- a/tests/testthat/test-sample_frac.r +++ /dev/null @@ -1,20 +0,0 @@ -context("test-sampe_frac") - -setup({ - a = data.frame(a = 1:100, b = 1:100) - - as.disk.frame(a, file.path(tempdir(), "tmp_sample_frac.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing sample_frac", { - a = disk.frame(file.path(tempdir(), "tmp_sample_frac.df")) - a40 <- sample_frac(a, 0.4) %>% collect - - expect_equal(nrow(a40), 40) - - expect_error(a40 <- sample_frac(a, 0.4, weight = 1) %>% collect) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_sample_frac.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-sample_n.r b/tests/testthat/test-sample_n.r deleted file mode 100644 index faf76259..00000000 --- a/tests/testthat/test-sample_n.r +++ /dev/null @@ -1,16 +0,0 @@ -context("test-sample_n") - -setup({ - a = data.frame(a = 1:100, b = 1:100) - - as.disk.frame(a, file.path(tempdir(), "tmp_sample_n.df"), nchunks = 5, overwrite = T) -}) - -test_that("testing semi_join where right is data.frame", { - a = disk.frame(file.path(tempdir(), "tmp_sample_n.df")) - expect_error(a40 <- sample_n(a, 40) %>% collect) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_sample_n.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-sas_to_csv.r b/tests/testthat/test-sas_to_csv.r deleted file mode 100644 index 0bdd6fa4..00000000 --- a/tests/testthat/test-sas_to_csv.r +++ /dev/null @@ -1,8 +0,0 @@ -# TODO everything - -context("test-sas_to_csv") - -test_that("testing sas_to_csv", { - # TODO tests - expect_equal(2L, 2L) -}) diff --git a/tests/testthat/test-sas_to_disk.frame.r b/tests/testthat/test-sas_to_disk.frame.r deleted file mode 100644 index dbd64e3e..00000000 --- a/tests/testthat/test-sas_to_disk.frame.r +++ /dev/null @@ -1,10 +0,0 @@ -# TODO everything - -context("test-sas_to_disk.frame") - - -test_that("testing sas_to_disk.frame", { - # TODO tests - expect_equal(2L, 2L) -}) - diff --git a/tests/testthat/test-semi_join.R b/tests/testthat/test-semi_join.R deleted file mode 100644 index 7b96c3ac..00000000 --- a/tests/testthat/test-semi_join.R +++ /dev/null @@ -1,72 +0,0 @@ -context("test-semi_join") - -setup({ - - a = data.frame(a = 1:100, b = 1:100) - b = data.frame(a = 51:150, b = 1:100) - d = data.frame(a = 1:50, b = 1:50) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_sj.df"), nchunks = 4, overwrite = TRUE) - as.disk.frame(b, file.path(tempdir(), "tmp_b_sj.df"), nchunks = 5, overwrite = TRUE) - as.disk.frame(d, file.path(tempdir(), "tmp_d_sj.df"), overwrite = TRUE) - - as.disk.frame(a, file.path(tempdir(), "tmp_a_sj2.df"), nchunks = 4, overwrite = TRUE) - as.disk.frame(b, file.path(tempdir(), "tmp_b_sj2.df"), nchunks = 5, overwrite = TRUE) - as.disk.frame(d, file.path(tempdir(), "tmp_d_sj2.df"), overwrite = TRUE) -}) - -test_that("testing semi_join where right is data.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_sj.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_sj.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_sj.df")) - bc = collect(b) - dc = collect(d) - - abc = semi_join(a, bc, by = "a") %>% collect - expect_equal(nrow(abc), 50) - - abc0 = semi_join(a, bc, by = c("a","b")) %>% collect - expect_equal(nrow(abc0), 0) - - abc100 = semi_join(a, bc, by = "b") %>% collect - expect_equal(nrow(abc100), 100) - - abd50 = semi_join(a, dc, by = "b") %>% collect - expect_equal(nrow(abd50), 50) -}) - -test_that("testing semi_join where right is disk.frame", { - a = disk.frame(file.path(tempdir(), "tmp_a_sj2.df")) - b = disk.frame(file.path(tempdir(), "tmp_b_sj2.df")) - d = disk.frame(file.path(tempdir(), "tmp_d_sj2.df")) - - expect_warning({ - ab = semi_join(a, b, by = "a", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab), 50) - - expect_warning({ - ab0 = semi_join(a, b, by = c("a","b"), merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab0), 0) - - expect_warning({ - ab100 = semi_join(a, b, by = "b", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ab100), 100) - - expect_warning({ - ad50 = semi_join(a, d, by = "b", merge_by_chunk_id = F) %>% collect - }) - expect_equal(nrow(ad50), 50) -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_a_sj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_sj.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_sj.df")) - - fs::dir_delete(file.path(tempdir(), "tmp_a_sj2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_b_sj2.df")) - fs::dir_delete(file.path(tempdir(), "tmp_d_sj2.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-setup.r b/tests/testthat/test-setup.r deleted file mode 100644 index 4d3473d1..00000000 --- a/tests/testthat/test-setup.r +++ /dev/null @@ -1,7 +0,0 @@ -context("test-setup") - -test_that("testing sas_to_disk.frame", { - setup_disk.frame(workers = 2) - a = future::nbrOfWorkers() - expect_equal(a, 2) -}) \ No newline at end of file diff --git a/tests/testthat/test-shard.r b/tests/testthat/test-shard.r deleted file mode 100644 index 0a2e28e5..00000000 --- a/tests/testthat/test-shard.r +++ /dev/null @@ -1,28 +0,0 @@ -context("test-shard") - -setup({ -}) - -test_that("testing shard data.frame", { - set.seed(1) - a = data.table(a = rep(1:10, 10), b = 1:100) - a = shard(a, "a", nchunks = 2, overwrite = TRUE, outdir=file.path(tempdir(), "tmp_shard.df")) - - expect_equal(nchunks(a), 2) - expect_equal(nrow(a), 100) - expect_equal(ncol(a), 2) - - a1 = unique(get_chunk(a,1)$a) - a2 = unique(get_chunk(a,2)$a) - expect_equal(length(intersect(a1, a2)), 0) - - a3 = shard(a, "a", nchunks = 4, overwrite = TRUE) - - expect_equal(nchunks(a3), 4) - expect_equal(nrow(a3), 100) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_shard.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-shardkey.r b/tests/testthat/test-shardkey.r deleted file mode 100644 index 846b81bb..00000000 --- a/tests/testthat/test-shardkey.r +++ /dev/null @@ -1,17 +0,0 @@ -context("test-shardkey") - -setup({ -}) - -test_that("testing shardkey", { - set.seed(1) - a = data.table(a = rep(1:10, 10), b = 1:100) - a = shard(a, "a", nchunks = 2, overwrite = TRUE, outdir=file.path(tempdir(), "tmp_shardkey.df")) - - expect_equal(shardkey(a), list(shardkey="a", shardchunks=2)) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_shardkey.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-srckeep.r b/tests/testthat/test-srckeep.r deleted file mode 100644 index 5b6fa370..00000000 --- a/tests/testthat/test-srckeep.r +++ /dev/null @@ -1,17 +0,0 @@ -context("test-keep") - -setup({ - b = data.frame(a = 51:150, b = 1:100) - as.disk.frame(b, file.path(tempdir(), "tmp_srckeep.df"), nchunks = 5, overwrite = TRUE) -}) - -test_that("testing srckeep", { - b = disk.frame(file.path(tempdir(), "tmp_srckeep.df")) - b1 = b %>% srckeep("a") - expect_equal(ncol(b1 %>% collect), 1) - expect_equal(colnames(b1 %>% collect), "a") -}) - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_srckeep.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-tbl_vars.r b/tests/testthat/test-tbl_vars.r deleted file mode 100644 index 5cd00ffa..00000000 --- a/tests/testthat/test-tbl_vars.r +++ /dev/null @@ -1,16 +0,0 @@ -context("test-tbl_vars") - -setup({ -}) - -test_that("testing tbl_vars", { - a = data.table(a = rep(1:10, 10), b = 1:100) - a = shard(a, "a", nchunks = 2, overwrite = TRUE, outdir=file.path(tempdir(), "tmp_tbl_vars.df")) - - expect_setequal(tbl_vars(a), c("a","b")) -}) - - -teardown({ - fs::dir_delete(file.path(tempdir(), "tmp_tbl_vars.df")) -}) \ No newline at end of file diff --git a/tests/testthat/test-util.r b/tests/testthat/test-util.r deleted file mode 100644 index a9db9f37..00000000 --- a/tests/testthat/test-util.r +++ /dev/null @@ -1,8 +0,0 @@ -context("test-util") - -test_that("testing evalparseglue", { - x = 2 - y = 3 - expect_equal(evalparseglue("{x}+{y}"), 5) -}) - diff --git a/tests/testthat/test-write_disk.frame.R b/tests/testthat/test-write_disk.frame.R deleted file mode 100644 index c5650c81..00000000 --- a/tests/testthat/test-write_disk.frame.R +++ /dev/null @@ -1,39 +0,0 @@ -context("test-write_disk.frame") - -test_that("as.disk.frame works", { - ROWS = 1e3+11 - - tmp_write_disk.frame = tempfile() - tmp_write_disk.frame2 = tempfile() - - df = disk.frame:::gen_datatable_synthetic(ROWS) - dfdf <- as.disk.frame(df, tmp_write_disk.frame, overwrite = TRUE, nchunks = 5) - - a = dfdf %>% cmap(~{ - .x[1,] - }) %>% write_disk.frame(outdir = tmp_write_disk.frame2, overwrite = T) - - expect_equal(nrow(a), 5) - - fs::dir_delete(tmp_write_disk.frame) - fs::dir_delete(tmp_write_disk.frame2) -}) - -test_that("as.disk.frame fails if data frame has list-columns", { - df <- tibble::tibble("a" = c(1,2,3), "b" = list("a", "b", "c")) - expect_error(as.disk.frame(df, file.path(tempdir(), "tmp_write_disk.frame"), overwrite = TRUE, nchunks = 6)) -}) - -test_that("write_disk.frame shard works", { - mtcars_df = as.disk.frame( - mtcars, - outdir = file.path(tempdir(), "mt_shard_by_cyl"), - shardby = c("cyl","vs"), - nchunks = 3, - overwrite = TRUE) - - res = mtcars_df %>% collect_list - expect_equal(length(res), 3) - testthat::expect_type(res, "list") - -}) diff --git a/tests/testthat/test-zip_to_disk.frame.r b/tests/testthat/test-zip_to_disk.frame.r deleted file mode 100644 index b2e4c31e..00000000 --- a/tests/testthat/test-zip_to_disk.frame.r +++ /dev/null @@ -1,7 +0,0 @@ -context("test-zip_to_disk.frame") - -# TODO do some testing -test_that("testing zip_to_disk.frame", { - expect_true(TRUE) -}) - diff --git a/utils/build_utils.R b/utils/build_utils.R index b158a0bf..33103cf0 100644 --- a/utils/build_utils.R +++ b/utils/build_utils.R @@ -73,19 +73,19 @@ df_setup_vignette <- function(excl = "", strip_number = FALSE) { df_test <- function() { # rename tests - if(fs::dir_exists("tests_manual")) { - fs::dir_copy("tests_manual", "tests") - Sys.sleep(3) # allow enough time for it to happen - fs::dir_delete("tests_manual") - } + # if(fs::dir_exists("tests_manual")) { + # fs::dir_copy("tests_manual", "tests") + # Sys.sleep(3) # allow enough time for it to happen + # fs::dir_delete("tests_manual") + # } devtools::test() - if(fs::dir_exists("tests")) { - fs::dir_copy("tests", "tests_manual") - Sys.sleep(8) # allow enough time for it to happen - fs::dir_delete("tests") - } + # if(fs::dir_exists("tests")) { + # fs::dir_copy("tests", "tests_manual") + # Sys.sleep(8) # allow enough time for it to happen + # fs::dir_delete("tests") + # } } df_build_vignettes_for_cran <- function() { From 01d247f0b36ae8c5bee3df8c3072ca191b2178ba Mon Sep 17 00:00:00 2001 From: xiaodaigh Date: Sun, 9 May 2021 20:14:52 +1000 Subject: [PATCH 7/8] ok --- .Rbuildignore | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/.Rbuildignore b/.Rbuildignore index c05278e2..5c5d5b95 100644 --- a/.Rbuildignore +++ b/.Rbuildignore @@ -64,4 +64,5 @@ vignettes.asis.template vignettes.Rnw.template ^codecov\.yml$ new-nse-dev.r -test-poorman.R \ No newline at end of file +test-poorman.R +*.parquet \ No newline at end of file From 9f1ebf6f6c5f416642b7daa8c417315d74d828bf Mon Sep 17 00:00:00 2001 From: ZJ Dai Date: Mon, 24 Jan 2022 21:40:22 +1100 Subject: [PATCH 8/8] updated the site --- README.Rmd | 2 +- README.md | 24 +- docs/404.html | 119 +-- docs/LICENSE-text.html | 113 +-- docs/articles/01-intro.html | 32 +- docs/articles/02-intro-disk-frame.html | 399 ++++---- docs/articles/03-concepts.html | 98 +- docs/articles/04-ingesting-data.html | 128 ++- docs/articles/05-data-table-syntax.html | 61 +- docs/articles/06-vs-dask-juliadb.html | 148 +-- .../figure-html/unnamed-chunk-2-1.png | Bin 28103 -> 28790 bytes .../figure-html/unnamed-chunk-3-1.png | Bin 27085 -> 27558 bytes docs/articles/07-glm.html | 104 +- docs/articles/08-more-epic.html | 128 +-- .../figure-html/unnamed-chunk-2-1.png | Bin 38731 -> 39178 bytes .../figure-html/unnamed-chunk-3-1.png | Bin 32146 -> 32544 bytes docs/articles/09-convenience-features.html | 56 +- docs/articles/10-group-by.html | 261 +++-- docs/articles/11-custom-group-by.html | 130 +-- docs/articles/88-trouble-shooting.html | 46 +- docs/articles/index.html | 159 +-- docs/authors.html | 153 +-- docs/index.html | 471 ++++----- docs/news/index.html | 287 ++---- docs/pkgdown.css | 83 +- docs/pkgdown.js | 4 +- docs/pkgdown.yml | 6 +- docs/reference/add_chunk.html | 279 ++---- docs/reference/as.data.frame.disk.frame.html | 273 ++---- docs/reference/as.data.table.disk.frame.html | 284 ++---- docs/reference/as.disk.frame.html | 224 ++--- docs/reference/chunk_group_by.html | 155 +-- docs/reference/cmap.html | 914 ++++++++---------- docs/reference/cmap2.html | 393 +++----- docs/reference/collect.html | 460 ++++----- docs/reference/colnames.html | 154 +-- docs/reference/compute.disk.frame.html | 198 ++-- docs/reference/create_chunk_mapper.html | 266 ++--- docs/reference/csv_to_disk.frame.html | 281 ++---- docs/reference/delete.html | 146 +-- docs/reference/df_ram_size.html | 137 +-- docs/reference/dfglm.html | 230 ++--- docs/reference/disk.frame.html | 189 ++-- docs/reference/dplyr_verbs.html | 218 ++--- docs/reference/evalparseglue.html | 140 +-- docs/reference/foverlaps.disk.frame.html | 238 ++--- docs/reference/gen_datatable_synthetic.html | 140 +-- docs/reference/get_chunk.html | 402 ++++---- docs/reference/get_chunk_ids.html | 201 ++-- docs/reference/group_by.html | 189 ++-- docs/reference/groups.disk.frame.html | 141 +-- docs/reference/hard_arrange.html | 281 ++---- docs/reference/hard_group_by.html | 292 ++---- docs/reference/head_tail.html | 200 ++-- docs/reference/index.html | 350 ++----- docs/reference/is_disk.frame.html | 158 +-- docs/reference/join.html | 445 ++++----- docs/reference/make_glm_streaming_fn.html | 287 ++---- docs/reference/merge.disk.frame.html | 214 ++-- docs/reference/move_to.html | 185 +--- docs/reference/nchunks.html | 190 ++-- docs/reference/ncol_nrow.html | 176 +--- docs/reference/one-stage-group-by-verbs.html | 208 ++-- docs/reference/overwrite_check.html | 162 +--- docs/reference/print.disk.frame.html | 142 +-- docs/reference/pull.disk.frame.html | 154 +-- docs/reference/rbindlist.disk.frame.html | 213 ++-- docs/reference/rechunk.html | 244 ++--- docs/reference/recommend_nchunks.html | 202 ++-- docs/reference/remove_chunk.html | 226 ++--- docs/reference/sample.html | 241 ++--- docs/reference/setup_disk.frame.html | 207 ++-- docs/reference/shard.html | 229 ++--- docs/reference/shardkey.html | 134 +-- docs/reference/shardkey_equal.html | 140 +-- docs/reference/show_ceremony.html | 127 +-- docs/reference/srckeep.html | 279 ++---- docs/reference/sub-.disk.frame.html | 195 ++-- docs/reference/tbl_vars.disk.frame.html | 142 +-- docs/reference/write_disk.frame.html | 257 ++--- docs/reference/zip_to_disk.frame.html | 237 ++--- docs/sitemap.xml | 294 ++++++ 82 files changed, 5816 insertions(+), 10259 deletions(-) create mode 100644 docs/sitemap.xml diff --git a/README.Rmd b/README.Rmd index 5b480669..e5d8beaf 100644 --- a/README.Rmd +++ b/README.Rmd @@ -250,7 +250,7 @@ ncol(flights.df) ## Hex logo -![disk.frame logo](inst/figures/logo.png?raw=true) +![disk.frame logo](inst/figures/logo.png) ## Contributors diff --git a/README.md b/README.md index 7f14e340..8b8b1af9 100644 --- a/README.md +++ b/README.md @@ -211,12 +211,12 @@ flights.df %>% filter(year == 2013) %>% mutate(origin_dest = paste0(origin, dest)) %>% head(2) -#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight -#> 1 2013 1 1 517 515 2 830 819 11 UA 1545 -#> 2 2013 1 1 533 529 4 850 830 20 UA 1714 -#> tailnum origin dest air_time distance hour minute time_hour origin_dest -#> 1 N14228 EWR IAH 227 1400 5 15 2013-01-01 05:00:00 EWRIAH -#> 2 N24211 LGA IAH 227 1416 5 29 2013-01-01 05:00:00 LGAIAH +#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier +#> 1 2013 1 1 517 515 2 830 819 11 UA +#> 2 2013 1 1 533 529 4 850 830 20 UA +#> flight tailnum origin dest air_time distance hour minute time_hour origin_dest +#> 1 1545 N14228 EWR IAH 227 1400 5 15 2013-01-01 05:00:00 EWRIAH +#> 2 1714 N24211 LGA IAH 227 1416 5 29 2013-01-01 05:00:00 LGAIAH ``` ### Group-by @@ -313,7 +313,7 @@ To find out where the disk.frame is stored on disk: ``` r # where is the disk.frame stored attr(flights.df, "path") -#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpEdhlRv\\file33ec67c31d9c.df" +#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpIlXNzn\\file568813b835a7.df" ``` A number of data.frame functions are implemented for disk.frame @@ -350,7 +350,7 @@ ncol(flights.df) ## Hex logo -![disk.frame logo](inst/figures/logo.png?raw=true) +![disk.frame logo](inst/figures/logo.png) ## Contributors @@ -444,11 +444,3 @@ ways? Here are some ways you can contribute [![](https://cranlogs.r-pkg.org/badges/disk.frame)](https://cran.r-project.org/package=disk.frame) [![](http://cranlogs.r-pkg.org/badges/grand-total/disk.frame)](https://cran.r-project.org/package=disk.frame) -[![Travis build -status](https://travis-ci.org/xiaodaigh/disk.frame.svg?branch=master)](https://travis-ci.org/xiaodaigh/disk.frame) -[![AppVeyor build -status](https://ci.appveyor.com/api/projects/status/github/xiaodaigh/disk.frame?branch=master&svg=true)](https://ci.appveyor.com/project/xiaodaigh/disk.frame) - -## Live Stream of `{disk.frame}` development - -- diff --git a/docs/404.html b/docs/404.html index ed268161..78100fa4 100644 --- a/docs/404.html +++ b/docs/404.html @@ -1,66 +1,27 @@ - - - - + + + + - Page not found (404) • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + - - - - -
                    -
                    - + +
                    + + + - - -
                    +

                    Page not found (404)

                    @@ -163,31 +118,31 @@

                    Page not found (404)

                    +
                    -
                    -
                    -

                    Developed by Dai ZJ.

                    +
                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    -
                    +
                    + + - - diff --git a/docs/LICENSE-text.html b/docs/LICENSE-text.html index a8f5414f..abba5cc8 100644 --- a/docs/LICENSE-text.html +++ b/docs/LICENSE-text.html @@ -1,66 +1,12 @@ - - - - - - - -License • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -License • disk.frame - + + - - - -
                    -
                    - -
                    - -
                    +

                    License

                    @@ -165,31 +96,27 @@

                    License

                    +
                    -
                    -
                    -

                    Developed by Dai ZJ.

                    +
                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    Site built with pkgdown 2.0.2.

                    -
                    -
                    +
                    - - + + diff --git a/docs/articles/01-intro.html b/docs/articles/01-intro.html index ce4dac15..584809c3 100644 --- a/docs/articles/01-intro.html +++ b/docs/articles/01-intro.html @@ -19,6 +19,8 @@ + +
                    +

                    Preface - The birth of disk.frame

                    -

                    ZJ

                    +

                    ZJ

                    - Source: vignettes/01-intro.Rmd + Source: vignettes/01-intro.Rmd
                    -
                    -

                    -The story of how {disk.frame} came to be

                    +
                    +

                    The story of how {disk.frame} came to be +

                    I was working at one of Australia’s biggest banks and their shiny new SAS server was experiencing huge instability issues. As a result, we had to run SAS on our laptops to perform huge amounts of data manipulation. A simple SQL query can take around 20 minutes.

                    I had enough.

                    That’s why I created disk.frame - a larger-than-RAM data manipulation framework for R. The same query now only takes 10 seconds.

                    @@ -147,11 +143,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -160,5 +158,7 @@

                    + + diff --git a/docs/articles/02-intro-disk-frame.html b/docs/articles/02-intro-disk-frame.html index dd9b0746..fec5669f 100644 --- a/docs/articles/02-intro-disk-frame.html +++ b/docs/articles/02-intro-disk-frame.html @@ -19,6 +19,8 @@ + +
                    +

                    Quick Start: Basic Operations with nycflights13

                    -

                    ZJ

                    +

                    ZJ

                    - Source: vignettes/02-intro-disk-frame.Rmd + Source: vignettes/02-intro-disk-frame.Rmd
                    -
                    -

                    -Quick Start - replicating dplyr’s tutorial on nycflight13

                    -

                    The disk.frame package aims to be the answer to the question: how do I manipulate structured tabular data that doesn’t fit into Random Access Memory (RAM)?

                    +
                    +

                    Quick Start - replicating dplyr’s tutorial on nycflight13 +

                    +

                    The disk.frame package aims to be the answer to the question: how do I manipulate structured tabular data that doesn’t fit into Random Access Memory (RAM)?

                    In a nutshell, disk.frame makes use of two simple ideas

                    1. split up a larger-than-RAM dataset into chunks and store each chunk in a separate file inside a folder and
                    2. provide a convenient API to manipulate these chunks

                    disk.frame performs a similar role to distributed systems such as Apache Spark, Python’s Dask, and Julia’s JuliaDB.jl for medium data which are datasets that are too large for RAM but not quite large enough to qualify as big data.

                    -

                    In this tutorial, we introduce disk.frame, address some common questions, and replicate the sparklyr data manipulation tutorial using disk.frame constructs.

                    -
                    -

                    -Installation

                    +

                    In this tutorial, we introduce disk.frame, address some common questions, and replicate the sparklyr data manipulation tutorial using disk.frame constructs.

                    +
                    +

                    Installation +

                    Simply run

                    -install.packages("disk.frame") # when CRAN ready
                    +install.packages("disk.frame") # when CRAN ready

                    or

                    -devtools::install_github("xiaodaigh/disk.frame")
                    +devtools::install_github("xiaodaigh/disk.frame")
                    -
                    -

                    -Set-up disk.frame -

                    +
                    +

                    Set-up disk.frame +

                    disk.frame works best if it can process multiple data chunks in parallel. The best way to set-up disk.frame so that each CPU core runs a background worker is by using

                     setup_disk.frame()
 
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
                    +options(future.globals.maxSize = Inf)

                    The setup_disk.frame() sets up background workers equal to the number of CPU cores; please note that, by default, hyper-threaded cores are counted as one not two.

                    -

                    Alternatively, one may specify the number of workers using setup_disk.frame(workers = n).

                    +

                    Alternatively, one may specify the number of workers using setup_disk.frame(workers = n).

                    -
                    -

                    -Basic Data Operations with disk.frame -

                    -

                    The disk.frame package provides convenient functions to convert data.frames and CSVs to disk.frames.

                    -
                    -

                    -Creating a disk.frame from data.frame +
                    +

                    Basic Data Operations with disk.frame

                    +

                    The disk.frame package provides convenient functions to convert data.frames and CSVs to disk.frames.

                    +
                    +

                    Creating a disk.frame from data.frame +

                    We convert a data.frame to disk.frame using the as.data.frame function.

                    -library(nycflights13)
-library(dplyr)
-library(disk.frame)
-library(data.table)
+library(nycflights13)
+library(dplyr)
+library(disk.frame)
+library(data.table)
 
 # convert the flights data to a disk.frame and store the disk.frame in the folder
 # "tmp_flights" and overwrite any content if needed
 flights.df <- as.disk.frame(
   flights, 
-  outdir = file.path(tempdir(), "tmp_flights.df"),
+  outdir = file.path(tempdir(), "tmp_flights.df"),
   overwrite = TRUE)
 flights.df

                    You should now see a folder called tmp_flights with some files in it, namely 1.fst, 2.fst…. where each fst files is one chunk of the disk.frame.

                    -
                    -

                    -Creating a disk.frame from CSV

                    +
                    +

                    Creating a disk.frame from CSV +

                    -library(nycflights13)
+library(nycflights13)
 # write a csv
-csv_path = file.path(tempdir(), "tmp_flights.csv")
-data.table::fwrite(flights, csv_path)
+csv_path = file.path(tempdir(), "tmp_flights.csv")
+data.table::fwrite(flights, csv_path)
 
 # load the csv into a disk.frame
-df_path = file.path(tempdir(), "tmp_flights.df")
+df_path = file.path(tempdir(), "tmp_flights.df")
 flights.df <- csv_to_disk.frame(
   csv_path, 
   outdir = df_path,
@@ -204,15 +197,15 @@ 
                    
 flights.df

                    If the CSV is too large to read in, then we can also use the in_chunk_size option to control how many rows to read in at once. For example to read in the data 100,000 rows at a time.

                    -library(nycflights13)
-library(disk.frame)
+library(nycflights13)
+library(disk.frame)
 
 # write a csv
-csv_path = file.path(tempdir(), "tmp_flights.csv")
+csv_path = file.path(tempdir(), "tmp_flights.csv")
 
-data.table::fwrite(flights, csv_path)
+data.table::fwrite(flights, csv_path)
 
-df_path = file.path(tempdir(), "tmp_flights.df")
+df_path = file.path(tempdir(), "tmp_flights.df")
 
 flights.df <- csv_to_disk.frame(
   csv_path, 
@@ -228,26 +221,26 @@ 
                    
 flights.df

                    disk.frame also has a function zip_to_disk.frame that can convert every CSV in a zip file to disk.frames.

                    -
                    -

                    -Simple dplyr verbs and lazy evaluation

                    +
                    +

                    Simple dplyr verbs and lazy evaluation +

                    -flights.df1 <- select(flights.df, year:day, arr_delay, dep_delay)
+flights.df1 <- select(flights.df, year:day, arr_delay, dep_delay)
 flights.df1
                    -class(flights.df1)
+class(flights.df1)
 #> [1] "disk.frame"        "disk.frame.folder"
                    -

                    The class of flights.df1 is also a disk.frame after the dplyr::select transformation. Also, disk.frame operations are by default (and where possible) lazy, meaning it doesn’t perform the operations right away. Instead, it waits until you call collect. Exceptions to this rule are the *_join operations which evaluated eagerly under certain conditions see Joins for disk.frame in-depth for details.

                    +

                    The class of flights.df1 is also a disk.frame after the dplyr::select transformation. Also, disk.frame operations are by default (and where possible) lazy, meaning it doesn’t perform the operations right away. Instead, it waits until you call collect. Exceptions to this rule are the *_join operations which evaluated eagerly under certain conditions see Joins for disk.frame in-depth for details.

                    For lazily constructed disk.frames (e.g. flights.df1). The function collect can be used to bring the results from disk into R, e.g.

                    -collect(flights.df1) %>% head(2)
+collect(flights.df1) %>% head(2)
 #>    year month day arr_delay dep_delay
 #> 1: 2013     1   1        11         2
 #> 2: 2013     1   1        20         4

                    Of course, for larger-than-RAM datasets, one wouldn’t call collect on the whole disk.frame (because why would you need disk.frame otherwise). More likely, one would call collect on a filtered dataset or one summarized with group_by.

                    Some examples of other dplyr verbs applied:

                    -filter(flights.df, dep_delay > 1000) %>% collect %>% head(2)
+filter(flights.df, dep_delay > 1000) %>% collect %>% head(2)
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   9      641            900      1301     1242           1530
 #> 2: 2013     1  10     1121           1635      1126     1239           1810
@@ -258,7 +251,7 @@ 
                    
 #> 1: 2013-01-09 14:00:00
 #> 2: 2013-01-10 21:00:00
                    -mutate(flights.df, speed = distance / air_time * 60) %>% collect %>% head(2)
+mutate(flights.df, speed = distance / air_time * 60) %>% collect %>% head(2)
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   1      517            515         2      830            819
 #> 2: 2013     1   1      533            529         4      850            830
@@ -269,13 +262,13 @@ 
                    
 #> 1: 2013-01-01 10:00:00 370.0441
 #> 2: 2013-01-01 10:00:00 374.2731
                    -
                    -

                    -Examples of NOT fully supported dplyr verbs

                    +
                    +

                    Examples of NOT fully supported dplyr verbs +

                    The chunk_arrange function arranges (sorts) each chunk but not the whole dataset. So use with caution. Similarly chunk_summarise creates summary variables within each chunk and hence also needs to be used with caution. In the Group By section, we demonstrate how to use summarise in the disk.frame context correctly with hard_group_bys.

                     # this only sorts within each chunk
-chunk_arrange(flights.df, dplyr::desc(dep_delay)) %>% collect %>% head(2)
+chunk_arrange(flights.df, dplyr::desc(dep_delay)) %>% collect %>% head(2)
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   9      641            900      1301     1242           1530
 #> 2: 2013     1  10     1121           1635      1126     1239           1810
@@ -286,7 +279,7 @@ 
                    
 #> 1: 2013-01-09 14:00:00
 #> 2: 2013-01-10 21:00:00
                    -chunk_summarize(flights.df, mean_dep_delay = mean(dep_delay, na.rm =T)) %>% collect
+chunk_summarize(flights.df, mean_dep_delay = mean(dep_delay, na.rm =T)) %>% collect
 #>    mean_dep_delay
 #> 1:       12.32700
 #> 2:       12.01291
@@ -295,19 +288,19 @@ 
                    
 #> 5:       12.67813
 #> 6:       12.05854
                    -
                    -

                    -Piping

                    +
                    +

                    Piping +

                    One can chain dplyr verbs together like with a data.frame

                    -c4 <- flights %>%
-  filter(month == 5, day == 17, carrier %in% c('UA', 'WN', 'AA', 'DL')) %>%
-  select(carrier, dep_delay, air_time, distance) %>%
-  mutate(air_time_hours = air_time / 60) %>%
-  collect %>%
-  arrange(carrier)# arrange should occur after `collect`
+c4 <- flights %>%
+  filter(month == 5, day == 17, carrier %in% c('UA', 'WN', 'AA', 'DL')) %>%
+  select(carrier, dep_delay, air_time, distance) %>%
+  mutate(air_time_hours = air_time / 60) %>%
+  collect %>%
+  arrange(carrier)# arrange should occur after `collect`
 
-c4  %>% head
+c4  %>% head
 #>    carrier dep_delay air_time distance air_time_hours
 #> 1:      AA        -7      142     1089       2.366667
 #> 2:      AA        -9      186     1389       3.100000
@@ -316,9 +309,9 @@ 
                    
 #> 5:      AA        -2      146     1085       2.433333
 #> 6:      AA        -7      119      733       1.983333
                    -
                    -

                    -List of supported dplyr verbs

                    +
                    +

                    List of supported dplyr verbs +

                     select
 rename
@@ -337,76 +330,76 @@ 
                    
 anit_join
                    -
                    -

                    -Sharding and distribution of chunks

                    +
                    +

                    Sharding and distribution of chunks +

                    Like other distributed data manipulation frameworks disk.frame utilizes the sharding concept to distribute the data into chunks. For example “to shard by cust_id” means that all rows with the same cust_id will be stored in the same chunk. This enables chunk_group_by by cust_id to produce the same results as non-chunked data.

                    The by variables that were used to shard the dataset are called the shardkeys. The sharding is performed by computing a deterministic hash on the shard keys (the by variables) for each row. The hash function produces an integer between 1 and n, where n is the number of chunks.

                    -
                    -

                    -Group-by

                    -

                    disk.frame implements the group_by operation some caveats. In the disk.frame framework, only a set functions are supported in summarize. However, the user can create more custom group-by functions can be defined.

                    +
                    +

                    Group-by +

                    +

                    disk.frame implements the group_by operation some caveats. In the disk.frame framework, only a set functions are supported in summarize. However, the user can create more custom group-by functions can be defined.

                    -flights.df %>%
-  group_by(carrier) %>% # notice that hard_group_by needs to be set
-  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
-  collect %>% 
-  arrange(carrier)
-#> # A tibble: 16 x 3
+flights.df %>%
+  group_by(carrier) %>% # notice that hard_group_by needs to be set
+  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
+  collect %>% 
+  arrange(carrier)
+#> # A tibble: 16 x 3
 #>    carrier count mean_dep_delay
-#>    <chr>   <int>          <dbl>
-#>  1 9E      18460          16.7 
-#>  2 AA      32729           8.59
-#>  3 AS        714           5.80
-#>  4 B6      54635          13.0 
-#>  5 DL      48110           9.26
-#>  6 EV      54173          20.0 
-#>  7 F9        685          20.2 
-#>  8 FL       3260          18.7 
-#>  9 HA        342           4.90
-#> 10 MQ      26397          10.6 
-#> 11 OO         32          12.6 
-#> 12 UA      58665          12.1 
-#> 13 US      20536           3.78
-#> 14 VX       5162          12.9 
-#> 15 WN      12275          17.7 
-#> 16 YV        601          19.0
                    +#> <chr> <int> <dbl> +#> 1 9E 18460 16.7 +#> 2 AA 32729 8.59 +#> 3 AS 714 5.80 +#> 4 B6 54635 13.0 +#> 5 DL 48110 9.26 +#> 6 EV 54173 20.0 +#> 7 F9 685 20.2 +#> 8 FL 3260 18.7 +#> 9 HA 342 4.90 +#> 10 MQ 26397 10.6 +#> 11 OO 32 12.6 +#> 12 UA 58665 12.1 +#> 13 US 20536 3.78 +#> 14 VX 5162 12.9 +#> 15 WN 12275 17.7 +#> 16 YV 601 19.0
                    -
                    -

                    -Restrict input columns for faster processing

                    +
                    +

                    Restrict input columns for faster processing +

                    One can restrict which input columns to load into memory for each chunk; this can significantly increase the speed of data processing. To restrict the input columns, use the srckeep function which only accepts column names as a string vector.

                    -flights.df %>%
-  srckeep(c("carrier","dep_delay")) %>%
-  group_by(carrier) %>% 
-  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
+flights.df %>%
+  srckeep(c("carrier","dep_delay")) %>%
+  group_by(carrier) %>% 
+  summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
   collect
-#> # A tibble: 16 x 3
+#> # A tibble: 16 x 3
 #>    carrier count mean_dep_delay
-#>    <chr>   <int>          <dbl>
-#>  1 9E      18460          16.7 
-#>  2 AA      32729           8.59
-#>  3 AS        714           5.80
-#>  4 B6      54635          13.0 
-#>  5 DL      48110           9.26
-#>  6 EV      54173          20.0 
-#>  7 F9        685          20.2 
-#>  8 FL       3260          18.7 
-#>  9 HA        342           4.90
-#> 10 MQ      26397          10.6 
-#> 11 OO         32          12.6 
-#> 12 UA      58665          12.1 
-#> 13 US      20536           3.78
-#> 14 VX       5162          12.9 
-#> 15 WN      12275          17.7 
-#> 16 YV        601          19.0
                    +#> <chr> <int> <dbl> +#> 1 9E 18460 16.7 +#> 2 AA 32729 8.59 +#> 3 AS 714 5.80 +#> 4 B6 54635 13.0 +#> 5 DL 48110 9.26 +#> 6 EV 54173 20.0 +#> 7 F9 685 20.2 +#> 8 FL 3260 18.7 +#> 9 HA 342 4.90 +#> 10 MQ 26397 10.6 +#> 11 OO 32 12.6 +#> 12 UA 58665 12.1 +#> 13 US 20536 3.78 +#> 14 VX 5162 12.9 +#> 15 WN 12275 17.7 +#> 16 YV 601 19.0

                    Input column restriction is one of the most critical efficiencies provided by disk.frame. Because the underlying format allows random access to columns (i.e. retrieve only the columns used for processing), hence one can drastically reduce the amount of data loaded into RAM for processing by keeping only those columns that are directly used to produce the results.

                    -
                    -

                    -Joins

                    +
                    +

                    Joins +

                    disk.frame supports many dplyr joins including:

                     left_join
@@ -420,11 +413,11 @@ 
                    
 
                    The below joins are performed lazily because airlines.dt is a data.table not a disk.frame:
                    
 
                     # make airlines a data.table
-airlines.dt <- data.table(airlines)
+airlines.dt <- data.table(airlines)
 # flights %>% left_join(airlines, by = "carrier") #
-flights.df %>% 
-  left_join(airlines.dt, by ="carrier") %>% 
-  collect %>% 
+flights.df %>% 
+  left_join(airlines.dt, by ="carrier") %>% 
+  collect %>% 
   head
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   1      517            515         2      830            819
@@ -448,9 +441,9 @@ 
                    
 #> 5: 2013-01-01 11:00:00   Delta Air Lines Inc.
 #> 6: 2013-01-01 10:00:00  United Air Lines Inc.
                    
 
                    -flights.df %>% 
-  left_join(airlines.dt, by = c("carrier")) %>% 
-  collect %>% 
+flights.df %>% 
+  left_join(airlines.dt, by = c("carrier")) %>% 
+  collect %>% 
   tail
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     9  30       NA           1842        NA       NA           2019
@@ -474,20 +467,20 @@ 
                    
 #> 5: 2013-09-30 15:00:00                Envoy Air
 #> 6: 2013-09-30 12:00:00                Envoy Air
                    -
                    -

                    -Window functions and arbitrary functions

                    -

                    disk.frame supports all data.frame operations, unlike Spark which can only perform those operations that Spark has implemented. Hence windowing functions like min_rank and rank are supported out of the box.

                    +
                    +

                    Window functions and arbitrary functions +

                    +

                    disk.frame supports all data.frame operations, unlike Spark which can only perform those operations that Spark has implemented. Hence windowing functions like min_rank and rank are supported out of the box.

                    For the following example, we will use the hard_group_by which performs a group-by and also reorganises the chunks so that all records with the same year, month, and day end up in the same chunk. This is typically not advised, as hard_group_by can be slow for large datasets.

                     # Find the most and least delayed flight each day
-bestworst <- flights.df %>%
-   srckeep(c("year","month","day", "dep_delay")) %>%
-   hard_group_by(c("year", "month", "day")) %>%
-   filter(dep_delay == min(dep_delay, na.rm = T) || dep_delay == max(dep_delay, na.rm = T)) %>%
+bestworst <- flights.df %>%
+   srckeep(c("year","month","day", "dep_delay")) %>%
+   hard_group_by(c("year", "month", "day")) %>%
+   filter(dep_delay == min(dep_delay, na.rm = T) || dep_delay == max(dep_delay, na.rm = T)) %>%
    collect
    
-bestworst %>% head
+bestworst %>% head
 #>    year month day dep_delay
 #> 1: 2013     2  21       301
 #> 2: 2013     2  21        -9
@@ -497,13 +490,13 @@ 
                    
 #> 6: 2013     2  21        10

                    another example

                    -ranked <- flights.df %>%
-  srckeep(c("year","month","day", "dep_delay")) %>%
-  hard_group_by(c("year", "month", "day")) %>%
-  filter(min_rank(desc(dep_delay)) <= 2 & dep_delay > 0) %>%
+ranked <- flights.df %>%
+  srckeep(c("year","month","day", "dep_delay")) %>%
+  hard_group_by(c("year", "month", "day")) %>%
+  filter(min_rank(desc(dep_delay)) <= 2 & dep_delay > 0) %>%
   collect
 
-ranked %>% head
+ranked %>% head
 #>    year month day dep_delay
 #> 1: 2013     1   9      1301
 #> 2: 2013     1   9       253
@@ -514,14 +507,14 @@ 
                    
 
                    one more example
                    
 
                     # Rank each flight within a daily
-ranked <- flights.df %>%
-  srckeep(c("year","month","day", "dep_delay")) %>%
-  chunk_group_by(year, month, day) %>%
-  select(dep_delay) %>%
-  mutate(rank = rank(desc(dep_delay))) %>%
+ranked <- flights.df %>%
+  srckeep(c("year","month","day", "dep_delay")) %>%
+  chunk_group_by(year, month, day) %>%
+  select(dep_delay) %>%
+  mutate(rank = rank(desc(dep_delay))) %>%
   collect
 
-ranked %>% head
+ranked %>% head
 #>    year month day dep_delay rank
 #> 1: 2013     1   1         2  313
 #> 2: 2013     1   1         4  276
@@ -530,13 +523,13 @@ 
                    
 #> 5: 2013     1   1        -6  742
 #> 6: 2013     1   1        -4  633
                    -
                    -

                    -Arbitrary by-chunk processing

                    -

                    One can apply arbitrary transformations to each chunk of the disk.frame by using the delayed function which evaluates lazily or the map.disk.frame(lazy = F) function which evaluates eagerly. For example to return the number of rows in each chunk

                    +
                    +

                    Arbitrary by-chunk processing +

                    +

                    One can apply arbitrary transformations to each chunk of the disk.frame by using the delayed function which evaluates lazily or the map.disk.frame(lazy = F) function which evaluates eagerly. For example to return the number of rows in each chunk

                    -flights.df1 <- delayed(flights.df, ~nrow(.x))
-collect_list(flights.df1) %>% head # returns number of rows for each data.frame in a list
+flights.df1 <- delayed(flights.df, ~nrow(.x))
+collect_list(flights.df1) %>% head # returns number of rows for each data.frame in a list
 #> [[1]]
 #> [1] 56131
 #> 
@@ -556,7 +549,7 @@ 
                    
 #> [1] 56121

                    and to do the same with map.disk.frame

                    -map(flights.df, ~nrow(.x), lazy = F) %>% head
+map(flights.df, ~nrow(.x), lazy = F) %>% head
 #> Warning in map.disk.frame(flights.df, ~nrow(.x), lazy = F): map(df, ...) where
 #> df is a disk.frame has been deprecated. Please use cmap(df,...) instead
 #> [[1]]
@@ -579,12 +572,12 @@ 
                    
 
                    The map function can also output the results to another disk.frame folder, e.g.
                    
 
                     # return the first 10 rows of each chunk
-flights.df2 <- map(flights.df, ~.x[1:10,], lazy = F, outdir = file.path(tempdir(), "tmp2"), overwrite = T)
+flights.df2 <- map(flights.df, ~.x[1:10,], lazy = F, outdir = file.path(tempdir(), "tmp2"), overwrite = T)
 #> Warning in map.disk.frame(flights.df, ~.x[1:10, ], lazy = F, outdir =
 #> file.path(tempdir(), : map(df, ...) where df is a disk.frame has been
 #> deprecated. Please use cmap(df,...) instead
 
-flights.df2 %>% head
+flights.df2 %>% head
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
 #> 1: 2013     1   1      517            515         2      830            819
 #> 2: 2013     1   1      533            529         4      850            830
@@ -606,39 +599,39 @@ 
                    
 #> 4: 2013-01-01 10:00:00
 #> 5: 2013-01-01 11:00:00
 #> 6: 2013-01-01 10:00:00
                    
-
                    Notice disk.frame supports the purrr syntax for defining a function using ~.
                    
+
                    Notice disk.frame supports the purrr syntax for defining a function using ~.
                    -
                    -

                    -Sampling

                    +
                    +

                    Sampling +

                    In the disk.frame framework, sampling a proportion of rows within each chunk can be performed using sample_frac.

                    -flights.df %>% sample_frac(0.01) %>% collect %>% head
+flights.df %>% sample_frac(0.01) %>% collect %>% head
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time
-#> 1: 2013     1  16     1700           1708        -8     2018           2023
-#> 2: 2013     1  13     1115           1115         0     1407           1425
-#> 3: 2013     2   5     2005           1935        30       NA           2220
-#> 4: 2013     1   3     1425           1429        -4     1711           1754
-#> 5: 2013     5  19     1453           1455        -2     1638           1638
-#> 6: 2013    12  20     1618           1600        18     1745           1719
+#> 1: 2013     5  10      554            600        -6      645            659
+#> 2: 2013     8  28      752            800        -8     1015           1022
+#> 3: 2013     1   3      955            958        -3     1120           1137
+#> 4: 2013     8  22     2157           2100        57        7           2323
+#> 5: 2013     5  14     1853           1900        -7     2003           2048
+#> 6: 2013     8  25     1550           1540        10     1747           1747
 #>    arr_delay carrier flight tailnum origin dest air_time distance hour minute
-#> 1:        -5      B6     15  N334JB    JFK  FLL      175     1069   17      8
-#> 2:       -18      AA   2099  N3AWAA    LGA  MIA      150     1096   11     15
-#> 3:        NA      EV   4204  N17196    EWR  OKC       NA     1325   19     35
-#> 4:       -43      UA    257  N508UA    JFK  SFO      326     2586   14     29
-#> 5:         0      9E   3318  N908XJ    JFK  BUF       50      301   14     55
-#> 6:        26      US   2187  N946UW    LGA  DCA       46      214   16      0
+#> 1:       -14      US   2161  N747UW    LGA  DCA       38      214    6      0
+#> 2:        -7      UA    561  N513UA    LGA  DEN      218     1620    8      0
+#> 3:       -17      UA    258  N831UA    LGA  ORD      124      733    9     58
+#> 4:        44      DL   1247  N914DE    LGA  ATL      102      762   21      0
+#> 5:       -45      EV   5038  N741EV    LGA  BHM      111      866   19      0
+#> 6:         0      9E   3648  N8940E    JFK  CMH       69      483   15     40
 #>              time_hour
-#> 1: 2013-01-16 22:00:00
-#> 2: 2013-01-13 16:00:00
-#> 3: 2013-02-06 00:00:00
-#> 4: 2013-01-03 19:00:00
-#> 5: 2013-05-19 18:00:00
-#> 6: 2013-12-20 21:00:00
                    +#> 1: 2013-05-10 10:00:00 +#> 2: 2013-08-28 12:00:00 +#> 3: 2013-01-03 14:00:00 +#> 4: 2013-08-23 01:00:00 +#> 5: 2013-05-14 23:00:00 +#> 6: 2013-08-25 19:00:00
                    -
                    -

                    -Writing Data

                    +
                    +

                    Writing Data +

                    One can output a disk.frame by using the write_disk.frame function. E.g.

                     write_disk.frame(flights.df, outdir="out")
                    @@ -658,11 +651,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -671,5 +666,7 @@

                    + + diff --git a/docs/articles/03-concepts.html b/docs/articles/03-concepts.html index a238713e..418a1757 100644 --- a/docs/articles/03-concepts.html +++ b/docs/articles/03-concepts.html @@ -19,6 +19,8 @@ + +
                    +

                    Key {disk.frame} concepts

                    -

                    ZJ

                    +

                    ZJ

                    - Source: vignettes/03-concepts.Rmd + Source: vignettes/03-concepts.Rmd
                    -
                    -

                    -Key {disk.frame} concepts

                    +
                    +

                    Key {disk.frame} concepts +

                    There are a number of concepts and terminologies that are useful to understand in order to use disk.frame effectively.

                    -
                    -

                    -What is a disk.frame and what are chunks?

                    -

                    A disk.frame is a folder containing fst files named “1.fst”, “2.fst”, “3.fst” etc. Each of the “.fst” file is called a chunk.

                    +
                    +

                    What is a disk.frame and what are chunks? +

                    +

                    A disk.frame is a folder containing fst files named “1.fst”, “2.fst”, “3.fst” etc. Each of the “.fst” file is called a chunk.

                    -
                    -

                    -Workers and parallelism

                    -

                    Parallelism in {disk.frame} is achieved using the {future} package. When performing many tasks, disk.frame uses multiple workers to perform the tasks in parallel. A worker is an R session.

                    -

                    It is recommended that you run the following immediately after library(disk.frame) to set-up multiple workers. For example:

                    +
                    +

                    Workers and parallelism +

                    +

                    Parallelism in {disk.frame} is achieved using the {future} package. When performing many tasks, disk.frame uses multiple workers to perform the tasks in parallel. A worker is an R session.

                    +

                    It is recommended that you run the following immediately after library(disk.frame) to set-up multiple workers. For example:

                    -library(disk.frame)
+library(disk.frame)
 #> Loading required package: dplyr
 #> 
 #> Attaching package: 'dplyr'
@@ -152,24 +148,24 @@ 
                    
 #> 
 #>     intersect, setdiff, setequal, union
 #> Loading required package: purrr
-#> Registered S3 method overwritten by 'pryr':
-#>   method      from
-#>   print.bytes Rcpp
 #> 
-#> ## Message from disk.frame:
-#> We have 1 workers to use with disk.frame.
-#> To change that, use setup_disk.frame(workers = n) or just setup_disk.frame() to use the defaults.
-#> 
-#> 
-#> It is recommended that you run the following immediately to set up disk.frame with multiple workers in order to parallelize your operations:
-#> 
-#> 
-#> ```r
-#> # this will set up disk.frame with multiple workers
-#> setup_disk.frame()
-#> # this will allow unlimited amount of data to be passed from worker to worker
-#> options(future.globals.maxSize = Inf)
-#> ```
+#> ## Message from disk.frame:
+#> We have 1 workers to use with disk.frame.
+#> To change that, use setup_disk.frame(workers = n) or just setup_disk.frame() to use the defaults.
+#> 
+#> 
+#> It is recommended that you run the following immediately to set up disk.frame with multiple workers in order to parallelize your operations:
+#> 
+#> 
+#> ```r
+#> # this will set up disk.frame with multiple workers
+#> setup_disk.frame()
+#> # this will allow unlimited amount of data to be passed from worker to worker
+#> options(future.globals.maxSize = Inf)
+#> ```
+#> 
+#> 
+#> 
 #> 
 #> Attaching package: 'disk.frame'
 #> The following objects are masked from 'package:purrr':
@@ -182,7 +178,7 @@ 
                    
 #> The number of workers available for disk.frame is 6
 
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
                    +options(future.globals.maxSize = Inf)

                    For example, suppose we wish to compute the number of rows for each chunk, we can clearly perform this simultaneously in parallel. The code to do that is

                     # use only one column is fastest
@@ -192,13 +188,13 @@ 
                    
 # use only one column is fastest
 srckeep(df, "first_col")[,.N, keep = "first_col"]

                    Say there are n chunks in df, and there are m workers. Then the first m chunks will run chunk[,.N] simultaneously.

                    -

                    To see how many workers are available for disk.frame to use, run

                    +

                    To see how many workers are available for disk.frame to use, run

                    -future::nbrOfWorkers()
                    +future::nbrOfWorkers()
                    -
                    -

                    -How {disk.frame} works

                    +
                    +

                    How {disk.frame} works +

                    When df %>% some_fn %>% collect is called. The some_fn is applied to each chunk of df. The collect will row-bind the results from some_fn(chunk)together if the returned value of some_fn is a data.frame, or it will return a list containing the results of some_fn.

                    The session that receives these results is called the main session. In general, we should try to minimize the amount of data passed from the worker sessions back to the main session, because passing data around can be slow.

                    Also, please note that there is no communication between the workers, except for workers passing data back to the main session.

                    @@ -217,11 +213,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -230,5 +228,7 @@

                    + + diff --git a/docs/articles/04-ingesting-data.html b/docs/articles/04-ingesting-data.html index f1abeeb6..eb86d829 100644 --- a/docs/articles/04-ingesting-data.html +++ b/docs/articles/04-ingesting-data.html @@ -19,6 +19,8 @@ + +
                    +

                    Ingesting Data

                    -

                    ZJ

                    +

                    ZJ

                    - Source: vignettes/04-ingesting-data.Rmd + Source: vignettes/04-ingesting-data.Rmd
                    -
                    -

                    -Ingesting Data

                    -

                    One of the most important tasks to perform before using the disk.frame package is to make some disk.frames! There are a few functions to help you do that. Before we do that, we set up the disk.frame as usual

                    +
                    +

                    Ingesting Data +

                    +

                    One of the most important tasks to perform before using the disk.frame package is to make some disk.frames! There are a few functions to help you do that. Before we do that, we set up the disk.frame as usual

                    Setting up

                    -library(disk.frame)
+library(disk.frame)
 
 # set-up disk.frame to use multiple workers
-if(interactive()) {
+if(interactive()) {
   setup_disk.frame()
   # highly recommended, however it is pun into interactive() for CRAN because
   # change user options are not allowed on CRAN
-  options(future.globals.maxSize = Inf)  
+  options(future.globals.maxSize = Inf)  
 } else {
   setup_disk.frame(2)
 }
                    -
                    -

                    -Convert a data.frame to disk.frame -

                    +
                    +

                    Convert a data.frame to disk.frame +

                    Firstly, there is as.disk.frame() which allows you to make a disk.frame from a data.frame, e.g.

                    -flights.df = as.disk.frame(nycflights13::flights)
                    -

                    will convert the nycflights13::flights data.frame to a disk.frame somewhere in tempdir(). To find out the location of the disk.frame use:

                    +flights.df = as.disk.frame(nycflights13::flights)
                    +

                    will convert the nycflights13::flights data.frame to a disk.frame somewhere in tempdir(). To find out the location of the disk.frame use:

                    -attr(flights.df, "path")
                    +attr(flights.df, "path")

                    You can also specify a location to output the disk.frame to using outdir

                    -flights.df = as.disk.frame(nycflights13::flights, outdir = "some/path.df")
                    +flights.df = as.disk.frame(nycflights13::flights, outdir = "some/path.df")

                    it is recommended that you use .df as the extension for a disk.frame, however this is not an enforced requirement.

                    However, one of the reasons for disk.frame to exist is to handle larger-than-RAM files, hence as.disk.frame is not all that useful because it can only convert data that can fit into RAM. disk.frame comes with a couple more ways to create disk.frame.

                    -
                    -

                    -Creating disk.frame from CSVs

                    +
                    +

                    Creating disk.frame from CSVs +

                    The function csv_to_disk.frame can convert CSV files to disk.frame. The most basic usage is

                     some.df = csv_to_disk.frame("some/path.csv", outdir = "some.df")

                    this will convert the CSV file "some/path.csv" to a disk.frame.

                    -
                    -

                    -Multiple CSV files

                    +
                    +

                    Multiple CSV files +

                    However, sometimes we have multiple CSV files that you want to read in and row-bind into one large disk.frame. You can do so by supplying a vector of file paths e.g. from the result of list.files

                    -some.df = csv_to_disk.frame(c("some/path/file1.csv", "some/path/file2.csv"))
+some.df = csv_to_disk.frame(c("some/path/file1.csv", "some/path/file2.csv"))
 
 # or
-some.df = csv_to_disk.frame(list.files("some/path"))
                    +some.df = csv_to_disk.frame(list.files("some/path"))
                    -
                    -

                    -Ingesting CSV files chunk-wise

                    -

                    The csv_to_disk.frame(path, ...) function reads the file located at path in full into RAM but sometimes the CSV file may be too large to read in one go, as that would require loading the whole file into RAM. In that case, you can read the files chunk-by-chunk by using the in_chunk_size argument which controls how many rows you read in per chunk

                    +
                    +

                    Ingesting CSV files chunk-wise +

                    +

                    The csv_to_disk.frame(path, ...) function reads the file located at path in full into RAM but sometimes the CSV file may be too large to read in one go, as that would require loading the whole file into RAM. In that case, you can read the files chunk-by-chunk by using the in_chunk_size argument which controls how many rows you read in per chunk

                     # to read in 1 million (=1e6) rows per chunk
 csv_to_disk.frame(path, in_chunk_size = 1e6)
                    -

                    When in_chunk_size is specified, the input file is split into many smaller files using bigreadr’s split file functions. This is generally the fastest way to ingest large CSVs, as the split files can be processed in parallel using all CPU cores. But the disk space requirement is doubled because the split files are as large as the original file. If you run out of disk space, then you must clean R’s temporary folder at tempdir() and choose another chunk_reader e.g. csv_to_disk.frame(..., chunk_reader = "LaF").

                    +

                    When in_chunk_size is specified, the input file is split into many smaller files using bigreadr’s split file functions. This is generally the fastest way to ingest large CSVs, as the split files can be processed in parallel using all CPU cores. But the disk space requirement is doubled because the split files are as large as the original file. If you run out of disk space, then you must clean R’s temporary folder at tempdir() and choose another chunk_reader e.g. csv_to_disk.frame(..., chunk_reader = "LaF").

                    -
                    -

                    -Sharding

                    +
                    +

                    Sharding +

                    One of the most important aspects of disk.frame is sharding. One can shard a disk.frame at read time by using the shardby

                     csv_to_disk.frame(path, shardby = "id")

                    In the above case, all rows with the same id values will end up in the same chunk.

                    -
                    -

                    -Just-in-time transformation

                    +
                    +

                    Just-in-time transformation +

                    Sometimes, one may wish to perform some transformation on the CSV before writing out to disk. One can use the inmapfn argument to do that. The inmapfn name comes from INput MAPping FuNction. The general usage pattern is as follows:

                    -csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
+csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
   some_transformation(chunk)
 })

                    As a contrived example, suppose you wish to convert a string into date at read time:

                    -df = data.frame(date_str = c("2019-01-02", "2019-01-02"))
+df = data.frame(date_str = c("2019-01-02", "2019-01-02"))
 
 # write the data.frame 
-write.csv(df, file.path(tempdir(), "df.csv"))
+write.csv(df, file.path(tempdir(), "df.csv"))
 
 
 # this would show that date_str is a string
-str(collect(csv_to_disk.frame(file.path(tempdir(), "df.csv")))$date_str)
+str(collect(csv_to_disk.frame(file.path(tempdir(), "df.csv")))$date_str)
 ## chr [1:2] "2019-01-02" "2019-01-02"
 
 # this would show that date_str is a string
-df = csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
+df = csv_to_disk.frame(file.path(tempdir(), "df.csv"), inmapfn = function(chunk) {
   # convert to date_str to date format and store as "date"
-  chunk[, date := as.Date(date_str, "%Y-%m-%d")]
+  chunk[, date := as.Date(date_str, "%Y-%m-%d")]
   chunk[, date_str:=NULL]
 })
 
-str(collect(df)$date)
+str(collect(df)$date)
 ## Date[1:2], format: "2019-01-02" "2019-01-02"
                    -
                    -

                    -Reading CSVs from zip files

                    +
                    +

                    Reading CSVs from zip files +

                    Often, CSV comes zipped in a zip files. You can use the zip_to_disk.frame to convert all CSVs within a zip file

                     zip_to_disk.frame(path_to_zip_file)

                    The arguments for zip_to_disk.frame are the same as csv_to_disk.frame’s.

                    -
                    -

                    -Using add_chunk -

                    +
                    +

                    Using add_chunk +

                    What if the method of converting to a disk.frame isn’t implemented in disk.frame yet? One can use some lower level constructs provided by disk.frame to create disk.frames. For example, the add_chunk function can be used to add more chunks to a disk.frame, e.g.

                     a.df = disk.frame() # create an empty disk.frame
@@ -245,7 +239,7 @@ 
                    
 
                     delimited_to_disk.frame <- function(file, outdir, ...) {
   res.df = disk.frame(outdir, ...)
-  readr::read_delim_chunked(file, callback = function(chunk) {
+  readr::read_delim_chunked(file, callback = function(chunk) {
     add_chunk(res.df, chunk)
   }, ...)
   
@@ -255,9 +249,9 @@ 
                    
 delimited_to_disk.frame(path, outdir = "some.df")
                    
 
                    The above code uses readr’s read_delim_chunked function to read file and call add_chunk. The problem with this approach is that is it sequential in nature and hence is not able to take advantage of parallelism.
                    -
                    -

                    -Exploiting the structure of a disk.frame

                    +
                    +

                    Exploiting the structure of a disk.frame +

                    Of course, a disk.frame is just a folder with many fst files named as 1.fst, 2.fst etc. So one can simply create these fst files and ensure they have the same variable names and put them in a folder.

                    @@ -274,11 +268,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -287,5 +283,7 @@

                    + + diff --git a/docs/articles/05-data-table-syntax.html b/docs/articles/05-data-table-syntax.html index 46a112aa..c59cc037 100644 --- a/docs/articles/05-data-table-syntax.html +++ b/docs/articles/05-data-table-syntax.html @@ -19,6 +19,8 @@ + +
                    +

                    Using data.table syntax with disk.frame

                    -

                    ZJ

                    +

                    ZJ

                    - Source: vignettes/05-data-table-syntax.Rmd + Source: vignettes/05-data-table-syntax.Rmd
                    -
                    -

                    -disk.frame supports data.table syntax

                    +
                    +

                    +disk.frame supports data.table syntax +

                    -library(disk.frame)
+library(disk.frame)
 
 # set-up disk.frame to use multiple workers
-if(interactive()) {
+if(interactive()) {
   setup_disk.frame()
   # highly recommended, however it is pun into interactive() for CRAN because
   # change user options are not allowed on CRAN
-  options(future.globals.maxSize = Inf)  
+  options(future.globals.maxSize = Inf)  
 } else {
   setup_disk.frame(2)
 }
 
 
-library(nycflights13)
+library(nycflights13)
 
 # create a disk.frame
-flights.df = as.disk.frame(nycflights13::flights, outdir = file.path(tempdir(),"flights13"), overwrite = TRUE)
                    +flights.df = as.disk.frame(nycflights13::flights, outdir = file.path(tempdir(),"flights13"), overwrite = TRUE)

                    In the following example, I will use the .N from the data.table package to count the unique combinations year and month within each chunk.

                    -library(data.table)
+library(data.table)
 #> 
 #> Attaching package: 'data.table'
 #> The following object is masked from 'package:purrr':
@@ -158,18 +155,18 @@ 
                    
 #> The following objects are masked from 'package:dplyr':
 #> 
 #>     between, first, last
-library(disk.frame)
+library(disk.frame)
 
-flights.df = disk.frame(file.path(tempdir(),"flights13"))
+flights.df = disk.frame(file.path(tempdir(),"flights13"))
 
-names(flights.df)
+names(flights.df)
 #>  [1] "year"           "month"          "day"            "dep_time"      
 #>  [5] "sched_dep_time" "dep_delay"      "arr_time"       "sched_arr_time"
 #>  [9] "arr_delay"      "carrier"        "flight"         "tailnum"       
 #> [13] "origin"         "dest"           "air_time"       "distance"      
 #> [17] "hour"           "minute"         "time_hour"
 
-flights.df[,.N, .(year, month), keep = c("year", "month")]
+flights.df[,.N, .(year, month), keep = c("year", "month")]
 #>     year month     N
 #>  1: 2013     1 27004
 #>  2: 2013    10 28889
@@ -191,10 +188,10 @@ 
                    
 
                    All data.table syntax are supported. However, disk.frame adds the ability to load only those columns required for the analysis using the keep = option. In the above analysis, only the year and month variables are required and hence keep = c("year", "month") was used.
                    
 
                    Alternatively, we can use the srckeep function to achieve the same, e.g.
                    
 
                    -srckeep(flights.df, c("year", "month"))[,.N, .(year, month)]
                    
-
                    
-
                    
-External variables are captured
                    
+srckeep(flights.df, c("year", "month"))[,.N, .(year, month)]
                    +
                    +

                    External variables are captured +

                    disk.frame sends the computation to background workers which are essentially distinct and separate R sessions. Typically, the variables that you have available in your current R session aren’t visible in the other R sessions, but disk.frame uses the future package’s variable detection abilities to figure out which variables are in use and then send them to the background workers so they have access to the variables as well. E.g.

                     y = 42 
@@ -219,11 +216,13 @@ 
                    
 
 
       
                    
-  
                    Developed by Dai ZJ.
                    
+  
                    
+
                    Developed by Dai ZJ.
                    
 
                    
 
 
                    
-  
                    Site built with pkgdown 1.6.1.
                    
+  
                    
+
                    Site built with pkgdown 2.0.2.
                    
 
                    
 
       
                    
@@ -232,5 +231,7 @@ 
                    
   
 
 
+  
+
   
 
diff --git a/docs/articles/06-vs-dask-juliadb.html b/docs/articles/06-vs-dask-juliadb.html
index 55ab6931..053e7c5e 100644
--- a/docs/articles/06-vs-dask-juliadb.html
+++ b/docs/articles/06-vs-dask-juliadb.html
@@ -19,6 +19,8 @@
 
 
 
+    
+
     
                    
       
                    
+      
                    
   
                    
     
                    
       
                    Benchmarks 1: disk.frame beats Dask! disk.frame beats JuliaDB! Anyone else wanna challenge?
                    
-                        
                    ZJ
                    
+                        
                    ZJ
                    
             
       
-      Source: vignettes/06-vs-dask-juliadb.Rmd
+      Source: vignettes/06-vs-dask-juliadb.Rmd
       
 
     
                    
 
     
     
-
                    
-
                    
-Intro - Benchmark 1
                    
+
                    
+
                    Intro - Benchmark 1
+
                    
 
                    This is the first in a series to benchmark the performance of disk.frame vs other medium-data tools. For Python, we will benchmark Dask, and for Julia, we will benchmark JuliaDB.jl. In the process, I will do a warts-and-all account of the tools I have tested.
                    
-
                    The title was ‘inspired’ by this post titled “R beats Python! R beats Julia! Anyone else wanna challenge R?”.
                    
+
                    The title was ‘inspired’ by this post titled “R beats Python! R beats Julia! Anyone else wanna challenge R?”.
                    
 
                    
-
                    
-
                    
-TL;DR
                    
+
                    
+
                    TL;DR
+
                    
 
                    For this simple benchmark, disk.frame is faster. But Dask has a more convenient syntax. JuliaDB.jl is not ready for prime time.
                    
-
                    
-
                    
-Timings
                    
+
                    
+
                    Timings
+
                    
 
                    Please note I have not tried to record the precise times over many runs, but I aim illustrate the magnitude of speed of the different packages
                    
 
                    -library(ggplot2)
+library(ggplot2)
 
-df = data.frame(
-  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
-  timing = c(17.9, 50, 76)
+df = data.frame(
+  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
+  timing = c(17.9, 50, 76)
 )
 
-ggplot(df) +
-  geom_bar(aes(x = tool, weight = timing), stat = "count") + 
-  ylab("seconds") +
-  ggtitle("Convert to desired format timing")
                    
+ggplot(df) +
+  geom_bar(aes(x = tool, weight = timing), stat = "count") + 
+  ylab("seconds") +
+  ggtitle("Convert to desired format timing")

                    Next up is timings for the simple aggregation

                    -df = data.frame(
-  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
-  timing = c(1, 1.2, 6)
+df = data.frame(
+  tool = c("disk.frame", "Dask", "JuliaDB.jl"),
+  timing = c(1, 1.2, 6)
 )
 
-library(ggplot2)
-ggplot(df) +
-  geom_bar(aes(x = tool, weight = timing), stat = "count") + 
-  ylab("seconds") +
-  ggtitle("Count(*) group-by timings")
                    +library(ggplot2) +ggplot(df) + + geom_bar(aes(x = tool, weight = timing), stat = "count") + + ylab("seconds") + + ggtitle("Count(*) group-by timings")

                    -
                    -

                    -Data

                    -

                    The data can be obtained from Rapids.ai’s Fannie Mae Data distribution page. I have downloaded the 17 Years data which contains dat on 37 million loans with over 1.89 billions rows in Performance datasets.

                    +
                    +

                    Data +

                    +

                    The data can be obtained from Rapids.ai’s Fannie Mae Data distribution page. I have downloaded the 17 Years data which contains dat on 37 million loans with over 1.89 billions rows in Performance datasets.

                    To download and the data, here are some examples

                    For Linux users

                    wget http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz
 tar xzvg mortgage_2000-2016.tgz

                    For Mac users

                    -
                    curl http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz -o mortgage_2000-2016.tgz
+
                    curl http://rapidsai-data.s3-website.us-east-2.amazonaws.com/notebook-mortgage-data/mortgage_2000-2016.tgz -o mortgage_2000-2016.tgz
 tar xzvg mortgage_2000-2016.tgz
                    
 
                    For Windows users, just download the file and extract the tgz file.
                    -
                    -

                    -Benchmark exercise: converting CSV to desired format and simple aggregation

                    +
                    +

                    Benchmark exercise: converting CSV to desired format and simple aggregation +

                    We find the largest possible single file to give each of the tool a test run.

                    -
                    -

                    -disk.frame

                    +
                    +
                    disk.frame +

                    First up is disk.frame

                    -suppressPackageStartupMessages(library(disk.frame))
-system.time(setup_disk.frame()) # ~4s
+suppressPackageStartupMessages(library(disk.frame))
+system.time(setup_disk.frame()) # ~4s
 #> The number of workers available for disk.frame is 6
 #>    user  system elapsed 
-#>    0.17    0.12    5.55
                    +#> 0.18 0.03 2.22

                    We note that there is some time needed for disk.frame to start up all the workers. Next we try to convert the largest CSV file to disk.frame format. The file to be converted is about 2.2GB in size

                    -time_to_convert_disk.frame = system.time(df1 <- csv_to_disk.frame("c:/data/Performance_2004Q3.txt", header = FALSE))[3]
+time_to_convert_disk.frame = system.time(df1 <- csv_to_disk.frame("c:/data/Performance_2004Q3.txt", header = FALSE))[3]
 
 time_to_convert_disk.frame
 #> elapsed 
-#>    41.8
                    +#> 28.77

                    Now that we have converted it, we want to a count by the first column. To achieve this we use a “two-stage” aggregation strategy. Note that use keep="V1" to bring only the column V1 into RAM. This avoids the reading of other unnecessary columns and should speed-up the analysis significantly

                    -time_to_agg_disk.frame = system.time(summ <- df1[,.N, V1, keep = "V1"][, .(N = sum(N)), V1])
+time_to_agg_disk.frame = system.time(summ <- df1[,.N, V1, keep = "V1"][, .(N = sum(N)), V1])
 
 time_to_agg_disk.frame
 #>    user  system elapsed 
-#>    0.07    0.11   14.30
                    +#> 0.13 0.03 7.89

                    We can inspect the result as well.

                     summ
@@ -225,18 +221,18 @@ 
                    
 #> 389490: 999994125744  15

                    Another way to perform the analysis is to use dplyr syntax to perform group-by in one-stage which is:

                    -system.time(df1 %>% 
-  srckeep("V1") %>% 
-  group_by(V1) %>% 
-  summarise(N = n()) %>% 
+system.time(df1 %>% 
+  srckeep("V1") %>% 
+  group_by(V1) %>% 
+  summarise(N = n()) %>% 
   collect)
 #>    user  system elapsed 
-#>    2.00    1.41    6.74
                    +#> 1.89 0.14 5.30

                    However, the dplyr syntax tends to be slightly slower than using data.table syntax. This may be improved as much of the overhead is due to inefficient use of NSE.

                    -
                    -

                    -Dask

                    +
                    +
                    Dask +

                    To test Dask

                    import dask.dataframe as dd
 from datetime import datetime
@@ -255,9 +251,9 @@ 
                    
 aa = a.value_counts().compute()
 print(datetime.now() - startTime)  # 1.2

                    The aggregation time is also longer, although in practice 1.2s. vs around 0.5s isn’t that big of a difference. We shall see how the differential changes when dealing with larger datasets in a future session.

                    -
                    -
                    -Dask is good
                    +
                    +
                    Dask is good +

                    A good feature of Dask is that you need not convert the data to parquet before doing the aggregation. This is a very convenient feature that is not available in disk.frame yet. For example:

                    startTime = datetime.now()
 a = dd.read_csv("c:/data/perf/Performance_2004Q3.txt", sep="|",
@@ -270,9 +266,9 @@ 
                    
 
                    However this strategy is not particularly fast. In practice, you are almost always better off spending some time to convert your data to a efficient format before performing analysis. This may have a time cost, but the effort will pay off very quickly.
                    -
                    -

                    -JuliaDB

                    +
                    +
                    JuliaDB +

                    I really like Julia but it’s a shame that JuliaDB is no where near as mature as either disk.frame nor Dask for inputting data. I have not been able to load the CSV using JuliaDB’s native methods and have resorted to using CSV.jl to read the data and convert it to JuliaDB.jl. Also, it’s difficult to figure out how to run JuliaDB from disk, and the memory usage is enormous compared to the disk.frame’s and Dask’s.

                    using JuliaDB, CSV, OnlineStats
 path = "c:/data/perf"
@@ -292,9 +288,9 @@ 
                    
 
                    @time a = JuliaDB.load("c:/data/.jldb") # 0s
                    -
                    -

                    -Conclusion

                    +
                    +

                    Conclusion +

                    It is somewhat surprising that disk.frame is the speed king in this benchmark study, but I can’t take any of the credit as the speediness of disk.frame is due to the authors of fst, future (and parallel), and data.table. It’s also no surprise that Julia’s medium-data tool lags behind Python’s and R’s, because it’s a pattern noticed elsewhere in the data ecosystem. Julia’s niche and strength at this point is in computational problems that require lots of computation that do not necessarily involved large amounts of input data (the Celeste project appears to be an exception but I think the computational demand there dominate the data demand).

                    Although disk.frame is the fastest, it’s syntax is not as convenient as Dask’s. Using Dask is almost the same as using pandas. In contrast, when using disk.frame, the user needs to be aware that operations happen in chunks, and hence a “two-stage” group-by is required. However, this will be addressed in a future planned package disk.frame.db which is will allow the user to “forget” the underlying architecture is made of chunks, and just focus on higher-level data operations.

                    @@ -312,11 +308,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -325,5 +323,7 @@

                    + + diff --git a/docs/articles/06-vs-dask-juliadb_files/figure-html/unnamed-chunk-2-1.png b/docs/articles/06-vs-dask-juliadb_files/figure-html/unnamed-chunk-2-1.png index 190fcde019b5c163b70d4c5901296dc9fafbc86b..663114824bffb3d69ceea54ac0ae84f49544dd9e 100644 GIT binary patch literal 28790 zcmeFa2UL{j+BH0C)I>#2G*$$n4x*r_ARt|1)C>qHRk})(uJn$^L}LMTXew0^kgine z7@`PBFGEMAccePB`S(30@9}@X^?l!a&RhPq{&ieA3yuym&vQTbeO>$7dtVnfPb;0| zShaN(g+k$={q&;>g|gh4LRohH>lOGLy`{DK__E@H!pR@;6@Qq=a=h^6n@c}w*-LJ_!kX<*EY(PMu{;W|;cntXTq z^ip#Dlq)~48(T`g`Og2!Ch|2R;;UTpm9m`g-E#7E_wFV8$=4%Czj0ndzF8&z54^?j z`6!h=oU)(J<$v_H=_+8&6{HSlv_yNc7Mo0ht!&?HztQ#hwovI#U+J1!;~#=Gl5`(D zeE6`guC8k;*~>#YsMbt1`qUnVjN9~3n2_l;bypUH!RUR>W%Ks@4er2l@eb)@H`bO1 zw@*g9kJyFBA4#9|SN!PS@tp;3YWI#MHk@XbmT&iEw$ktIJZUo1=2W?<;M1%59`S;b zpO>#rF)4qzSNFvMr*;?9=u;2qdX7zeZbQ#cF3vIvrtO>NS%E@&4^t~;-TVgQW9sVb zqmq(@rlzK3^yFhi4;}I!6bg`+(W^KSTep~;oLsqN<+i)N?p-^xI(?*4ESpmH8stm+ z$}Kt%3=9aU?W1)Y%y3U-?g<+WzMzOzJWhQ#yhI^zZ;9vJa8ReOob2jNLZ|4K&tGkn zn5t8?d$mE+s5#9nYqCaR;wqzwX~7D@fFDJ9Kv;A2gUvo_?jC#rOhuW)DZe%E^%{q*Fe1*rJFQZpz zQU2l8?bI@%oGxZ>xH8=`v(1tGKq_l-E+#H6POC+v;_%Mo+;RB|*|{NIhkA7yy~K~- z^v&Jv64Q-_IeCtB^>~eLk#OeQU5rg?zfbnpmem%_ckj1rJbz>AoM@hEl)+5U~OYxR~L(cDl6NsmwhPMCg0__c^6A0v(8jy>dr5S@_Et!7oY&X1ToRCBULQ9;S$K7s+p~4J zz=+hZa@(F{WQeK8DD9IA6SLBeDQcs}#>JI#dVMMy@5H+AyS;H=+(mD^m~!O#*UmX5 zceaSsR8`%jZaH}2$LS6)*3^jMXx>c4i_4wkHEsUCe{P1d6i1q7!_wAyhqOq~DG!G9Cx>5Dx<4kPC zu!ZTC{nu7L8>v}CrS@BFvt9^LESMvL>tL>QGLA`g6tu1NLu?fAp zY^BNEaAsxoN#4--r0h893I`n=K9%;4RMSf9g9WL*CoGENR9fz@b#UNCp5qDUO86Qt85*bSXpT^#IJ7i3a&QDbCudlC9Z!Xp{4OO?!be$Tg!PPZe8|YX*KW^5Z>#}dw?Bd0X z8fDo2k_jxtN`cI9?9bawUzzE%We{$kgunXfoLOiCMx*bzZ%FWr2M$^hrbBN_LeN2Z%&g{lA^#OO?@P+w-nJ2NT zF&Q}{`@NWDQ7W-&aq;nj+_=}_jL?IICvxUu^YinII-iepTv@VAds4Gx<}OEna^yrF z|Mc)|L)xS8qlyl7#ducom9le|{dKC!`{XorbXxj55y-37&t4bO6bKN;BOdSkX8kVv z{#l`{F1q{s)dsuectp+Blq9 zy{57g4|F}I>S^?%wNiUWT^TvT&SUT1J`@!shidEAtts~XHA8tb?JkSV-qh|uoxq^s z+8E^uiQW+R#ffm`6i4Tq$~bjaN%7C09}Lp9r7=I^@m58tszXOnAgjF8Y{Dg zTk=q)9LwFhU}5@gsFXO(_T#HfF}S-9WJz&nIqGmYm)>)4TK9~Xj*ryHi*faFl1v&B zwEU>dmz9r`LseAc2Q=oGuw^v3h?c$m1 z99#_-^nf8oT`Jv5i-l;iV^=x53YT%deRjSm1)-tUu2LooA<{2_yX=$RsH^58v(n4t z2Djr+Dd&^trq!|wqOYKpX1PldSNpS<>DYPpIbllo4OZ*BbO)Gdel3Hvrl#xL_Vj%A)bv`SPUumWol8#Zyg{iFy2av)f3J z@IZpGzHHpXFZ=Q4#$rRfVU(UzTY76xsbI3DLmmlbS>;d9M9S~a9s41Ec%zW+*=Dnt z&{8XY<*?SrihLc%8e$5jva~aDriZWBkyAd??q=iS>EPhdUm-PWS>4keWFW-t{O;<$ zjLHqborTd#g)JNpuLslQZ*ST^LZ3P$CKi~1^t-m`mO~g2jfsQI$3NDW_Y};QP2m|b zObiqYqe)KLCqw~ zX}?+ak`BLI$-8%E0?;(C%~MHgIIRYNMuB&s@1#Pd^u(##%)#|zjy%<&iEsiGm zENxFZ%*!y9bDEYhLpm>S_U4t#FP9T2e<&K?Is#zk6>s(0m)F30Vy=Ug-)6N~#JhOf zItRfi?Y@Ekl}xE=t0;4UfM(Vjg!jF9?>TICSK8Pd_7Syc@NdWYTn$8i<%i3D>DJFi zu8R(;X=*x^xbk>@dM`ckR)p5DXf%sJYL&TwxN19+)f>nwz98`_p4+AC-WCMS;E7^< zc9`dfoBY~VsUxOSG~0zKMrO0g)IfdKNcJF|T6CSGEX-|0D=ZVcW`4+24giWb!=W?A z;l!QIXLwd!cy((-79MGf73-d_`;#)C^6LYohRuOq2$t;AL~R00x9xj$BcsW&@X8Wq z`%FH=+ABcXdhj~uPYr8toHp-`R@2mTQ12-Z4#IvGO7>b9Y|%44DJdx~kmViF&I+ij z@9mt;UEzzTDAwVZ(QF=X3AEB(OkiVh(S}^q6?>KCa5DGx0BIS2_~Yh-!ZOF?+$2kT48+DD!TWHI-sBHE8@5=Y>{(-y<5V!$HasnYBx2&r z#l@wQ=29%xeun4T#CX~M{T+$9u{k5z@wr2!=rCy+yCFCKNZJGF*!W%H)!_zzD)0Qe zm1VVqOJ8Ih;_wsSpyzU`#g?@&6BMJ2e73guqjcBZW6hxkC7lh2q=O~wOxi#CI2j@I z#wI2XKdbbvuFg7BVwEttnEWJ3mY2~h>Rh;FWtx;1z|tONaSoQg6#;(z(OyT+#uyZM zW(}p330pKINH2cs?p)(E<&;vk)*swnFZ8E)6>(77yrS$IY~5m zt*drUN=kI}u9$F7*ZjG*QIq;O^#@W;*1MEK8WR+2gX29u74ce&jov%SvqzcH(9q+x z*z47txQDhlommhN6MX$rV$*jvakT-(Y5O>($BwwTwv}f#i&!>3BN4A$PFHRG1;^Gy zy8)_{n9({UmyG1kjZjNQaz?{QJ++e8{ZjjEL#D=|i2lCTr(I$it}IOz&2ZnriLI8d z4i_yR@~|Eq!^TV!+3w)>7Z^tQ;kNjxSgxlh*rxsIqS*4_(tIm(Y?+6P)lWo{WAi=c zOj=)Dejsks?yB((pos`7hX|CjTO@3an7!G`vP>gK#K2q0rfrqkem$CJ(@nAQ%XyM@ zvr_8Ey6cC<>WbwSKf2H4bSTnwFTdJQzO;Rlf%`!qEO|PC^7}k}=DPfp17o}q-?ejBocb0X~`=`Cmb7Yt~69@lx^&}d+2z>`nimqToYUBCZ{A2F_reEqRd@I;?M5~emo}b zGV0c$6E<>jZ{u{UZP`AF_rI*SZL!Mi&gu=b5KGSc_*2XHfN}FI^YTeKg^>huS6ieU zE$Ll|?Pa$%2$oI|@Em3>Y27QC$T9a!R#H1(rR6M*oQRs{sx90uQwOMuxI@6E?%!AM zf4h2r?p-{qxh5QPoxQ4>nn48ZG+89)+5Y|wzh#wJo@?l!YUiwZwtLi!7oWov^M!1$ zxzpk=`Dy~Bgf6Libj%LmxCi*59F?oiFFCkeJr~RF9WCG8;BUM?(;HoM%<_%I8VcoE zETXy}ZZ}EVqs@U~H5T-i4gUed3A8dP+CsG``qhDyLrPPbw%w&BzB!!&4Zy>JF{W?t z-!mEx8gY;VR7J}{WlFIk?p7SdqFm|^yDT(qo0oTu?sQfxGC%wOV~t9Wkqoc5tpu^;ZA$>_5&eR%^!jW1b3=mOxq~QSAYq z+hNo@<)BDHTqb&e{Par-7N?SR-9NsP(cPEaVBhC^{Pso%^9MMg*|+0B?L3u6;3Gv1 zlRsMV#*VYR$}&@Gi}Ss5LA+k`oxC<;UZ^>CD~odrrphQ5{Nb}zRKR1-ap5U9@2A3* zrL{qNj#~6|CmDm8iBVI$e5UvepGDY8;WdK$_LkvFI4IVm_etwfrbcUZB+N6rareSP>JkLC@12 zbtWm+MNBK+YGSKz0}k0D=jqP8e&KlDb8VXphQ3=s46ke7qqQyKVq!f=+C!yu6z&4@ zg)NG+Q-hlMKZt6%nKJ9+HJx0UDzW}Fzg+>m1;bW076*MskFMla3(KD!h?km5{TMC& zmE`!Fol<(0uQv-D{p2G%+eapHbL3{PjBd-=18 zoF1#OShgWnTj2d$k;=Wwl_3&r9FC_c-{I05s-tI8_w-B?7zR2Ob#Y&v@}vDZLn%Yg ziq=X@RvnuFYPl3SXzhCJk%Js#q%`04!+latY!SEC%_@jK+8+D=$su0%1>=N^2t9j*N&5YFJj&L zV*aDAm*qrnC-8RZO1V!jO}3YCdX8Ni_bz=nfHKcIt2=PK6MLZ>&(bP=l;pQ^Piv2^ zdnW^rMU6ZI&@ly4rqOGar1k-xDy2BOw@+I~+tn(H2gJlEg+4^}d6eYoNma)y&Vr{Xmj46sf35RX!}3AS9J1Gxz1qTrKlw z>yraJ;`Oaw{=A%EAd<(#+!pJ5j{t~gKQ3s0svIs`zPR}^Sl6_}N+r#~&qM8NXjw@1 z*{RYX&b&1eUjEJP>OsTTUD#)(s3k&;%L2Qx2g?Y>b0+dc_oq(ZN>E&cQ6U)f0c!iy zct^i3x`I$iSBYQCkXIXb`~K{{$5e}D&k>{#`0*=O5PEj6e(qtQp#8SJp-zE8E(iq; zdruFn^_ZP3!Nv+jX~qvWbdQ`@-IB ztrYR?K8hnzg1cR^P*+B1HKsiRZ8?5jJa;>ZT=XtN$|xOv|I6fPhaAYZKU~}T^HMz! ziMX*dh4LnA+Xs){9H$=|+T^UA{^8*~_`dsRAz6 zZ1`~o+qxV3zJhFy4T2g*_y_{UK{|1EM_gN53U*{1*}n((zx&3|$F8rTb>w@LjatNO zCJXUKK1mtfu2*efGiAVeF;KfsOIC7x%1;?QQ*F^{GX_@Ncr&)AR%Us^=iN8nP%{_v|L_;tnh9wZb-VYS2{MJN0TLo+HBW|sdbS{2%ue?E5ppne)ZLUE&Yt%-cq^81O5gq1S?rX} z0)6k%E!giPxA=*0?;rq_Ga4`4&h6wLeS^9re{?G;aq4n(COV@>= z{B^~Nad85q^g^wFKR!8M5=_h46TUp5vL3TWspVp2cwVH`>;Vs`04Sbsna}Lm(6p&7 z%X?6b0Pt^d0*pZzl|Mdy+ZJf5oAAP8?~8Er;deG4+;M}~(-~PRW$)Q%-v=3Zx`4l+ z&w$YEHYO=&BNB+r6T#U*Tt=L*ajC!bOsiei*zc=TJ3Qy2P;TJimSUB|T5JkrIAtbJ z6Jpz>Y{SAlv@F+xIF(rMmj+8H`i2_G`k_mft)S8f0R{a1{DvUto{Qmv&P%T7TeJm? zAZ}StyuPC-L&)!b!Z{k7;rwiPzsrJhBv}-7h%{243rbsCgbcEF@~9SrhZ${u9Hhy9 zddcZ{@+Mx-^$N5GUVpCe3VE(S(f8_P>@}GGu&K7Hs_L;G0vULudt5fT2Wi3=tG*{v z+v%@gLwR#e$HQTxdz2=AYz1UbOb9*Hsa*mY1{M6taS->1*x~O17^$9N8!tEI0KnBM zGe~<8j9?tVwK9`cCWb>9CX;>DY*`9;%uyAJB!%lFM7-Qxhme>v3l^-b!Tl4(faq0V zZbcl28n=O!kr;RyDSY|EACAR+Sii=)um4m4OBEaN9{->e_)n{ev(xrV^4i9`%W4}N z8*7azT<6E$y`<5vU%#G!qUuKcl8866Go#_4c=>4N78VO~-(KO+(9lePlp{A}LfY(uJv1;-te>gS2I&zN#cw|0{ zC_J;Xv-s&NJ9K6qsAy{T=B)Zom$kSc?yPm}5PK2+o0QXkdb$2f_1pi=Rm9mSv(;j~ z#7B>|vnA0LYNrq>H4?Q0s9Ndf>NsC*MH`NGgOZA|F?eol+N6rbnsg;{O;YZ7X_(Nm zD@ykPQV2U2CKP#KO*MNl|EF~6|IK~54Y?q z?Mc9qdx-YE;`9;}jmq=~ik38^6_o{W*)QZ#F^1YIIm)LCLl6Jm?pyHOj zCSobR5?y85ehX!C9i;o={&nv}jUg$^=q;s$))jcM7_Bp-+1c3~`Wl5TuW+38UanrA zIkMUryNl59JV`6A?BF_l=^W5T5LN5>h2xNITwm;P9)1c^tbZM)?+MsZqox#N8NCyk z104md75o1@?u(skmp4qtqxLsgFgWP;Y^$?X(|y56N}qo6 ztI-Zfv0S+cFZZ&yiy=_1k!o=I*%nnjo@)Ph0J;L89Z&0rZT&SwLdVRSLd+Sz7aMm zwGA5*pz8aT?GP&56|1>CmtZtP*>;0HcG{)*(4QpIl4G1Q{(DrTyMA6R^d`-$wzM)x zK8pPrc-L_Vbh7LDl}@FI?OtqV*9m$q4K&?4+TxYL9cwlp z_@0!ao&f5-1Q8SK2hn^_^0*kS;N)qwd_WiinlK6>jk#@blF-wO ziNgN8x}NM!fa^q>x2li>V@@8d^&68V!T0gB2tr3Fryw@UehZ_CWAe(r6QoFD2UEDQ zNJRE#dYw=H+Ietr(2q*WKLXL+`xOjIvHczx4OcL+v4XJ}dwav&tRcE*Bk_y3m?e5Q z11b*)da^Zvx#br6m8~k0zc$btQ@DVqro{t|)1v0<{oQ z#)_AVecDm0|Lxdxe=R?;?Fc64(t$)Xnoatu9?@#{hHQ~>wI^55+}y0OtR1B>9}UG_ zeNQ$%v`pO4q8v8{+*uVnOc9e&ym1Mc7ACf#m*$k++0jM6h>D5Levx3q^+kRx(svsU z=wkiQq^T1$E1p-GK`9XRq2AV{T#7odjI zBgclQw1g4t?uCdkhq$r}ex3^WB2+y=8pskmfgcqWw03Iv#m@&zAWmM;_#l99n>y(TN6sRx=FpRKVS*W3iigpJCE`P|SAKE9v zX3|`jWh7=a3xIgGVxM(TXXQkmU{jPhE~H8!|TGV}`>`H5iBa>5OX|l{` zfDcFxs>KMNHv-Ayl9M~$886S(vjo2=qxVB5N`f-*q*bWwtG01FeYDcYcp@lr& zcy?NE&R`5e4J=9EXntn+Fq)l2uIz}1nY1)O&@KCpb@~YFH?B0x<%1C$z>+E*J9bPt zsM<=bCz*p{pA07?f&f9RvtyF z?d~U5VA^<2$3oJV9~p{FOz*m~Z0*JA`RwOTGcYC-zr_6fn|nL`s052@`m@tZm7oKy zFbB*Sm8~#T=3z|GKyZZt3)3oN; zNSZLvQD43q@=FI+2E{=rjv*FS7xn}ieP2f74mLlZn3!0OHw>amK-$fDyp7!U75!L+ zrR{*}e6$gm;f7vx`ZNJxh}iYKrIysis-mz7phBn95r$ySb~({~Jtrge&a5RreI&V# zZ^bUqrLShkwrApiTP+7v5_y@d;Fktop* zWIbF-8L-X1>Zru0ch^)AI;C|v5 zyG8uF3QrPy8(f*HH3GbJrZ~-2bucuQ1_BIT2 zlDCb>v|avZ7Z|24l#FOn3(F5ql}@ z0tWOhVs1ECye;+M=RK7hZK+n+qVatWd|MFc!+)ad+LrJ!AwC8pZkt#^nDr$F(yXZl z9Xp~3z#d-SD?5|ry@!1w-ibyjT)+4U5`8$4zad}P;2$f~zjg|i?744fODs}=({O;# zWPkZMiG7}C3pnZ&w*Y%!AeDT;-Mt!0CYMjc}H*Al69cCMhSi6 zoCzMES_09kDR$addi*E45rl`nFJHlEbGO2|1XQ-nw1Rb6cCM*%7fp8lJ*U%g3HBS| zL4B1AYvl!cm#^2nAyEd=qmf(h2rajQhz2KlJr2=qYiRrq|Mn9dNqzQEEGuBo4j`rv zcG!Wa)%)%mX9a=X=q(W@KMxs91gLOaI7r{sj9}rU-Jhd~<^UaV5A6sx#>jvVgpHwR zq7;7%Cco*i9y9~%+>fufdCh;+Qz<3|FY2{J_(Xm3Ff>}ga0>%ZqeSiv+p9OrdhRrO zMuRc6SZv^CviX-KSzmwcyz{tbyHAkXxM#_aH9t#!Fn(Y^`NZr(;ICRu$qA7z-;4G? z)uOvFR3_**4((ebeKGds_5>t5Mt98loVSWfaOg-hgk3W`&bY+$(EkKfR-tz*Iz> z%%*6fA%uvIHT!8a`uh*BYWJCj3tO0c8q}Zs29GeWMJ6$U<79Q1f18<|ExAee(ouRB z%n42#*k|w=O<-o(=n?-vRMe?*x>x#G}P8YG*Y`~WP1=9oc=akuAt419ce5t8fX2S85_HYgm2He zmX5&A_l$lNf)yZd?CRAuqf%0AL+2`j#rM-{a-2prGCAR%qorJ|e#{|7C0A6q;&fo) z+--yT$9yAed1a+Li+wBCN)6q`ttL<0YMyo_Hl$8(@t!AsN~PcyCX(jh+uKEpi_>O_ zk;hFNhi@Wr;vl*?QW}QIK_haeQ)pBwO6b16Zz&feZ{50us@r~Yb5&ICgomjwk4yve zYGL8^^zEL8-(B&J#3Gc1#;d6%uzpx={7o)eDU|6{Y<|-)=MOi5LxVcaG@t_Azb%x# zn>2nQOt6yX3hZRwr62EAw<%<7*av^bL#Tfqi7mxsDpTo10uH1yXS`T5-s_zFx?lpVU@eZa&^eRj z!id_^Yz_-Fjc(nRZFRSpokkn)macQw)Z>;yt!XPgzqO|#G#KoMAn`2_uc}xZh^Q!- z?_k+f1T>A`o{eIdRs*6tUA6*hzc6}n(ycpC-=@ONE>>Q*BthnTlqjeMJ8F@#93V^Bisqo9>;Zr?&yG zvo9(Bdw;&Mu*KeepnDtpjv`bA%C_dx*j`cM5G(88lp1=r4c@(`3%B=; z55y<;pizt%EBvV80G_G90i@;dRMzXyINKiQ&j9^qA-1K!Q< ze%nwLe0<|`Lq969Uj*(q$gl9wOgMWF)&wf;^_?xMUw%As`Nlqd`i78?Bm)yiQvBRL zc(8cU;zwvxYDtXjRIKft&--eRdO+vVIVftxkuK4BM?@GKbPW;&?`+i#;@>wRZB1ww z6@gelj1T)}$fiKfQ~UB`sS*Iss>tSE3k)wsx-^`Vq^8hrn?b|KTXa#WCY||fjrLYQ zD`FVubhOL?tjHo)-grZMD3eZ!U* zFwump%>p)o7vprrLG~;0@RqleJPYep3EP;8GRML)T+~8C-?cv~#HHr}v5Jvi8(Z(s zwdEe1AyDYSO*RD!q`#6{f?G*>Eh`7;X$SWdpNrHAUYIGc%l3ZxEEUDu{P`3DpO{vL zF|k7f0b!D!R}x<$VLaK5VptEh!37-%Wd}c3I5BF7_AtOwX>iitN2vD-T0_@_VkOTWBc z$RQ+f{N=lxM_RCUC?K|EB4T*>2n z&U?XiW+ar@2a9>!EVE904kiAj8=d`(;B-YsEdNs&Be9kAq0LSY2NQ^TlG{NIpy-)B zf@X-}pBHP)0ce=0lerC@H>8#=TTo;Fi=MoetPzZ2*tz3^ZtwKq^d-G!a2+lNoo)Tg z3lLrcMTc0n(k*jvt?X8y)}4R7wcg|R#EXe$zaSuHjZ9?(e)xvV$f5*WXYWy0H#RuP z-iQ&kyIVHolI(WEXIs-cC!`04=(GzQi*za}A(SNhpZve}EUbDG*gAA7iH+E5J-<3U9u{NDgq;srv`E)5c*3 z&i(R(2xJ8>Pr@V7FKlu@o`|0VODRayNh}a~#srIZZ~k1`sR}5(OGHxL@A>gIwS*Y^ zNP7shDVru)Bw5m0Nw6!ygw(9}*STuoar?K3i>R@ygbXo&dDIAYI~tui(VJj+A+Hlu zP0kDHlP$AC;v(j5^zYv#hhkxFrt%y9kh@^Dg6zIy?{}pb%d0cswN627wi_?u?vIg|ni+H|Yf_oz zkFB~36Cr5yRxj4Vzt5^R+5ORV3`Ptv9iW}k2+C<{39!6Z4BnMS?1Jck5|S8BtF^_c z*hRBO*#7c)$RgI$=%b!9Iizh(+d5p*!MrSd28DQ_97+{Q+W4RWEBnhG18?^!IzcP( zZTaTIZO7J1P}{N{EV4Y<4mAY0fG^`ExI`Kq^+sR|87x6=g=ULR(ZmcdKs`E11Yho~ z&t#hy(v`Ng+iZV~&Z0Y-8nNO$^73-@N3(_C>2ioy zI@5_D%+cREMJ+igH(UMXU6Zy-8Xb$ioAeSnc6iK41ESOD#J!EiTQ)W`CbnqOb3-M) z#$VReh1x&pdUseaJTRDO6U);NodWT8;n zgH!faS804;7ubcfgH?Z$ZL{nDV)bFH#Gl=}ul_abbiW)vP@+9S*%vd%D7ei{94amX5)m-dm#Bc)1&BDB_5$ zHncnhFUSzPz$SV}eZ2!IuqQ{_q}pDS{EV}^xJlDh- zkHnXM&4c-qZ2abPt+kx*uemUv7ZJw1f4u;E=U?mapWY8@>L)p$Jz7Bfb0r}>VZfI} zkx8m>LfyGfM)E3FhRNvURYFI60LOt8!8?P={|gi8JRmlwm8QhzKW^uB|&LHeRK5Uq|i@JcV83i{eA zex?b27-0x6U>W7$WNAO*$3_g6n?rYvC6z&awFA`Qr;&Eoz?NP#a2rka%_+gkAWz?m zkpqdoc`#ngL&DM?oF=9xSY%u)MK-B}24lw|bu~58GB8L&q}G(y-%v0Atgh~GGqdv< z?P*fd2s@zRA%UJ#5eR6Q%h&tB-AxE)GFgcLaaceZwr(WFBusfbV9)qblcA|;^X5Q> zfT{nZHnE}%>kSMTT-_9LSU^A^yN+}FdDYZ`yj{;O!xJBgKpWwm?J(HDj;eK%`?U^57fmH!nxvb=bgZo4(3ZU3&F1V-;*zXLw1JcEy>{xxg^IDp##8|`_=*&o)j?fNg63M}f zzFdEl@evh|!{pbLv@=+yASq3=${M{;;W07L!E%e8a^pbQWC9cUw1!U`UZ8mqzV|5H zFEDxH>kgRJsWf0N#;h(;;mKT-bn9f)Gx&d5VFt913Aa8_EN9{B%4XAxa7691X%E_O zFp{_4@<+4_jsBHJ*rX`m9?7rh#nY+2P9a^Xis z#k{BT8Cf67a8>NmQ84iAxl6gYR@4*>H|rKfvqisPeRow0=m2cOi!7x2>H+>ZuSog z=fL5xjYq@+Rp~jM3c7C8l`W4!8!jVNomF#MV2fsw?gJFuyJ*CDBqpegP_IZOxC3XSqK5^Zn{{gP<+0X=f9LJ-7AxZW6>SgwhV2z;xiN&F9$A@Rh6|uO4gj0aTlCF0g z(6p(sJYu)&^iUR!S2MdgYs-;)h{;xdveEv>j=!2UxG|Uz`&K|uuw_c_ZxnykOzSpD z*IyiL*DPntH;S__KHRuYd3EN@nddZK)zU|?v1WOrgI{Z!z!t5DzJZgz&#OtR5IfpK zSy9DcscOT7T0|%|R=)`4FnIDZ#``6)UqZ-L_8Xyo9^_8f+sHo9|3ZHMY2@esuXiAc z=DYvvIq!&5s(}&w4NRY$} zzb})$?JecAC%5MQAHZl#R#sNz2@I8_P|n6u%|YfOxPUY)Ir*-FUluFXdJw(pH>#t*79-Yf4LQMMCZNlTA|7&~Rjf zsc$RP#nPrj=Ij$k>*UZWq)1O+18;o=njVPMc!?SHy?S zF4TcR`dcz>5J{9!TyV=QrC8NLMMvnVf$5MP&<5J`=5~3U+H+(2P`VJ8bB9Y*fAcD4 z+?b$th8|$n;&8`eB64U1$F0997ek3cJ3Z=jO4>_MBT@)w&ET9u;wj(#K}OG1@BB_a zh$GkPIVel2R`0si@B@&kn~ZFLo)N8lcjk(4E7gp84IO~RXGa}I+hqVcH*2FGYNhfr zu7+sOJ7lB*2#DHh!@?D{U@q0vau*J6#{7n7xf?<&qTO{N5M4_8Ax7oQnd%xr;w{>` z!hU1$_`#@c&}Mkx>>w>x-~plGrMYfUWph+stVIit5Qu2K09rlTnQ#C^BW$)!Eak=n zNdKz+_3IO?uq79R$Ya~4iN31PJIkK66No*2Z49x0uU>ZAHU6c|kunP3s9P3FtTF|) zq%`ieeTH`*MzNN(@cOf>lV`F0HCLSg=jG|S*p!LZ;h5#H?lW)Gc*vo@kf;$<8b&{6 zzs|^PcNHX597-7C82@scIbmW0>D3@zF3=zoyQc#LHBJ7KXXEv5f7T)W6=Jh)d)bd- zqY^C^l!+;(D3xHy+&HN4C}{R%B8z6hY`x?1cCsd7V%CK>uX@L-2^d2f`)Pcd(@P&R zU$C&{BrmR`NG>bhKoRqI;!)kE9PtUfJh8`2WhWP|ezq37_zju2 zOWh*t?o4Klu-82Yjo=f++C#I&NDgE0iX-oCQg+9&`ElZMqmrRR`IyY=H1^|>lg~f% z-hT%w|5wUO-Q3#RdZceVMNA!h0JK0;I+aW&bp(t1`N@M<9>;WIDh)<}8!*MBVy{O$ zC3G*yU_Kf##1Jfvzm>f8S27H|NP^=3ft8Gli(E;$n9MNnl7X&l9^V2eb&m`zCB0cN zcA|)GvFz$^pC*$d&?7eZ7V4hjg|;Q}DCA&uHUdHn*X-h>#F7kx2E%- zJ|?yoASw~koJJcK;F0Dm3i8zI^Uj?uhjp#t7-JAhZfR ztO4MWq|ny+L8x$-b^YZCWTd^m>5nSEZMIWNGY@Pa#~HQ~G7g4RI_{XJqm2HeW@$!t z{xrwOyyai%N319I2;`xRW+HD8wUEqD=^-6gq?-*TJV_6O5qp+2ZYE4lD_mvVGAC}M zQr$%(9Z&419kb+|>S-TCb+^i!WU>JIhH|?{!4#O8o0p=mhfF*$Mo=cE*t}-h(H=~) zY<=QWQnf|v+NQok+BpK5&`C*Mi=jni+zCI8IR8ilYX z!qh1>fdC?&f&}QsohcyEkm+5}mHm5HQyOjyI$sqs1ffxm%F&jXw=h`bO;G-16qG@{ z^Ek1}kUl^%37QC{WJeXtPCsA6E9>sPFgP$^3k!k?>;*&wC0;Z#qXDKiCCqAWcZeqA zVKAJ91`dYEdici+*W42+q1CNfc6Rv59%6e&iB6{Pkh!XfBRLrk94&L%p}Coi5ftDV zx_QA=gl58i5jG9G*92oyiMaqij;vPuTE%{(5?p{EnXSEV#$|tTG<3Qh#AXQ|B7{5_ z*e%O2y?}UwA;KLXLvRr;h!)=p`iI?`hM?a==Br>C2>Cw1I0G{oWcB@URKEdzL8g-T zkU2)gFsfs|w9mHPYN1ixqKM3y9@I^C)Uz1NXH3vD%*!OIn*l3`m5DHis*of>zvU{< zSXgoBpDTiCw<^lL?-;K<{S?;7J3T2WdmxU0EMi+F-AA16@0MZm0O^agZFlJ<+aZ3~ z04yFED@0^5KPnlhMVLb(z!JfQ4Dm>|3&g;fE;4+E$k>>MqSLzk>v zsGP&?`H^X=F5ot`+X~YET<#ghNGO=PbUzksUEiX0mqrcv|3 zZ}S<3wQ}QA0jaTQ24{?N9WfoZ9xMiRHJ{BBlh?~PTAy8U zvw_H06Fuc*#t-Imx46p<4*$SpUY3HOG29{!K?;7)^I0UWgi7VNVEJ znUW=piotMS_mb%^0`?6F+-CbO2)pxPJ2kqLhbgi6bk1!`5ggbJy5kS z5W$#mR}%eC{Kz=FCx{7%ZtPLQ#2&{MXhC$wcVBLYeTi%?ge>$$=KGOWKI7N$fKtie zrQ@6Q$ZRvKNzMby`|dPPGNx2uf!X`ZBa%d`DK zE5ZjzrJ>KK)pibtSp?~UOxV{MR#amPp8arSAgwj0T)#j^--8s545v@bU=bS+8E%dX zN?iOgH!Dj5t-f!uTZ7jt9e&i^#e1b;0jXGF+;N-U1}DEKutsLR`+2pBfk8)fu~r9T z_(^VZ8(A=gR>eh|g+Hbw#1b(XB_i-eaPERdI`3R#G1^-(H42fXF{R>yv9U3+b#)EL zJ9PAufrW&Z#<|S=qfK3e2ks+@zKCm>B{a{6O=U5fx#UXPXnH|ngQJo^73~-XY!g6t z0cKtq;XJTMv|&?s4W(9AkP@c7lEeQ-;c9z{_=cC-S+>SyR>UzvepvuqdWp}yuuBo% z{EX7qvxps^&CZ*G>Wv@vf;pm>%=_EQtMUve9>yxUH;7UquoV&$0oTyr(!K-@y@grd z^J26zTMnz|vdJwpra?}J%c~)&M6y>Jl7_U?SpYiEUS~kt-&JDX5kzHHvgRuT=h95z zgrv3SuiA2D+M?2e{CLleC13 zZ~_M<`OSwnpF!S6r2gE+gHDS5#OpkcF6taTI|`Q%w%nfhhWl|(pPt>fQ5u5-nHU@x z05OF0XONDVRFrVt8R{GUV_NN4tS- zIvcj|@F=~=8|2Q6y2+K}mnR0>Mp0_joi;$#;|3m68-kxM`AO>eYX)o6{x?%|3pBZ2 z?>h@OZ#gOZ{|0)1M)%>BOFDPp{#ej;ipp))j})_Rl=A^pXidM^sBsikAz$vMVVZK? zLs5$~qmp$2{{C;?U;0Vq6Mx)C*`I!GD6-9IYtHw`(G?1!#^&TmJ=S+|$r2V3K^y+w zeL5MZu3ft}i)>u(^@wtSc6i%biEV53a-9$MNM5a}=Qy9F`M5LV&_Tn@EsSU`a0XE! zA=|ute1wl}Dtsb(;6Q{om-qlA4PVKPvr7-6{CO<>o%!okwzw{sfBXi@%D2uLwyozI zU@dD?)AOR}%3;ulS<&E>IkFG^2|&#CvKu#U6j=yv+_2%5^!~K|Q+>zYj*UHoIVlZT z?CGpnu;S{Cd+$Cu=p{P^7t!vWJC7k>zUMTjT)Bu)NxyI-SwC45(*QPSa)`XP>hJMx zE71*Bd6bf(LZ1YxeI=dm#qtDOAe8$p59Mt2{(|{+c6%OH&3<_G>eaX1fQP}n|rHCo6-(VPMXh1Ge6i1Ww9bd8HUt2?+^Fr;!kdEfGh)|GwzJ zBpy&aVTc~Teo^H*j4rjU+qSIb(!)|shR8gb}5Wqsck^b$RC-Lpu#M0wGd8gVA7n#N$Llo zM1R7;CAM2mPL8jVlM+jAOh+dkO*2+48Hz0#R<{o?V#(tpBEEmTcmIB6p&$irZf;iu z(DaMdJC-k9GBr8*OK~r}UDWy@F77$MJpNV8DTcw$3h^~N%}>~(nBjq;>Zx6M?%Jg< zWXVnnIP(49K02ehD;pL;b#3i9G-{~}=}5~Wtd!G;nxo^icfq%m10`eXGlIC(724GxZ)}R}&#Ve{t4g0kW;6ec#;>rI=RhObyt!Pw!q!VHv~2+JKvR z5*WCVTh3EAGBQ$N-@bYnym;ExhO}|P59Pe_;Zozmy;xv}Xcu_qAu_?TA8K##b~9PL zTTY%PCef=_uNJjwm(FUl#&Z#)> zP;T$|EMMZE{)un*miA7RSZ9`5XAZ*Ow!V!UZ*z*mo{MyD4ZMMw$aCH|M^5N^78u~a z>*r6|YvLsbmau2oT{%2G+L8S8&r64)NI%XYzJ@=3Wb-k9TpnsXxcu77n?PAZ7hGMWd2sMi zF#fcXlSfTa(R+MiqQ13Nx@Y3S=QmmZ`Axh_OXa6#X5!=G;7{?GwI|;f6c`{0y>|su z-2f^g;oO0z7#?lbTM;@iJY0__nP$NdQVbSTyn5saLSl&HaPx+Vthem>9Q|b`x>Cyt z8-In`kIYby-l#)Jf}zqb!XFyny^FwgY*@2m3OhSKGV)kcQ`5Kq&r03g+@2KeEO~N5 zFf}FR#Otr8UY5+d)^`4Ra9-Xzm6MM@X1u(#bY+D!(993=$9O=K)(V`!>WPc3dD4<= z*RHKd2%EHN^X9Gj_w77`3?xE;ndfCLNYP$M67$T+&Yr9Ss)a#gS$C>hTU&t*qzNip zw``el4;-M<;INp`0xU>9F9A;sJiazMEDU@IpwV8??DP_Ia0+=~CE2!c;zU8uAVeAr8LDvkmO)q+h_|CnP9HC~#Zp>o8DhAzu%y zyQ6@Ut~1^Orvg?0OZs;oQ*M9wS##>_*{LO^z4w5df%|TP!e3234O|u!iinH91`bRX z?Tp!U`|YG9z!`!LtL*FRre1#?T3fppc>2wvH*a#DK6$cY(IO{c>Xwp_kTAYux*8ls v>JT%4#x=MaLPoZMY@II187RYt`KtbjjoiMZ{mqYoeq!)+^>bP0l+XkKALC+x literal 28103 zcmeIb2UOJewmm+^s3|IGY7~(q4x%6u3n1OZSOx?X1f(0KNLPA?#H%JzM299IU_(Tj zQl+Ct5fD(2Iur{}@fcft;n)~zp*ZTe5y6?Sv*IU2e%UyTbVCFlYa?U<`?{oNl zxAu+&pDg=?!C)+4?fglX!I5Cc7pytE2QhS6@&2&gZ0x7d%eyNwE3RbcP~<(Y5YFORR6=zPhFX>sX41kvtqVv z;%Q@Rx7SV)wr3XG8eRQ3uC^l7*w!m~m#UQQvNG+E5v?Dk7b)&nI9qsShhF9vc^|u+ z_}N-8^R!HDWA(9F$ zwEOc{AJFf<3JVhc&4mje77G7mj>OYB^qbYIXUWp9TYmW1YZm?H6OGbO>DS%6KcLUd z2>SJlf8ZsQ_a*JEI2fWD;T*{4@EQZB8wc|H)0&+M1~V$x6?9&^e7eFa(kwzU-Y?^F z<=+fajkBfN3OscthT2BD6-x3hSK8)&{N<{!{nxHq5E5hokSt6X#~;&g^g+%V^W7M^t*^XcF>3l=QM8>v>_c<9lEfr8;; zUY}t}w1r@@F-qCfsUgiQhBwmFQ_LT#{M*N0W=ypE+Ti6X;{{Var`N>@I@l7LF1P0_ zad7o_56jQbe^Os>)t2urW~Nb4-`E)0zb;%u%&dBG(vumBTqm=I$^$q2xS{b6izeD* z5;P;Nn=`Du$D3{R3=Iu0+Oebr`5pTLInQ~;G7TSML4W$7ec=tS*Uy)`Jor|zD@<=g z;LH@W{Iv4Wxy))M=|*#3$5lFU=abWmmBgoC?x?JM7^kYda{H+V!3yn7d#pmo3{3+C z{EnKSS+iX@j=7E-{eO^XJ$!dTl%lIO`$|rZ{EM=X8pqtW2=S?>^Ax8?>y&eDXM z_Z?n|rEwd4`9b@k#WKcWcv$81LaqkeGPlj6lD=W9i0UReIXSJCINzZ>qhderb+?}W zY8p5_#$q3NaX-Piqgb)`N@$MHAT!x8y?ROUt7DF7ve*8$M(=7^V&LnO7My>Kbg}f_)%N)E4q9`~kQJ)(`2&q5Gh?PM< z9@lmTPsMALm>16A?Y{r%uEA*RuwUZxp@mB~Txome#P%Dsx2Sye>~3g^cuC_TlPw9U z=0*7`ZJ9PT3sbNcj6<@sh7B@Yx+->^-+Xh*i`!W~xX?57%GIl@)F&T@RD=xUY1i(% zap97j@cP{Q?rEFGRqfd5$;nA(dFd~+U0+^2;p8GdZ-v+X<09g5H38$9ynC_%XWNT> zbDN9P($?@v*&w^AMttVg2^aMf%UyOBaB{a;XPJw z=vuAVV;O1YQC_V)d|oGj8yY3=Xlz{-9X`}nkci#xva;~*3g>3vHJm5Ro{D>So~;_P z*9__HKBwSpUOYYM+;idRLXCJe|BO0q<*u#ble?Yrx*`lhyiKHzw6ZN{rg(YX(e^G+ z_Uy3giS^X3iBlbk$SstRJb(Ddw33-g?&ikSsPBLN(L1o_-1^JSQ{!=>GGfUZjt6yl z(*mF1iM!A5ovf^x=&Mhjn#wI6OI;k9`X}$!8h+ku-`e97bv<#u>%NrKUOU`#aq9`| zfu@_h7vh2@2eS~>@qKCnUX9JmM;CjWGi_*Ne*aR^Y1Gtqpx)4;idJ!B_2!kngV}~Y z6GO4sE1MeCiHAvvana?Wk~a7LSX33kVK@3~%4e4Z#}13#bXG{XD#M%aQ?%{m;x69~ zzWMgxW#NOl9ct5)W0fb=r=IPK+K^BfSK~kA+U0Ke+#R`Mx4uNqAL|d^4`bth>k41p zxt>1U?mH|qHr!ECSluw=KizL3TVG$F-Bjk?6Km*{i*>KX)3sS!C=cB_Xgfa8Ts4`L zp6;w$*_!97U@SnmG;mV)=5|_-#Cvx~)!z8|u6(|~Lc7<&q>C3X){mV(a(`)8j^It1 zy2k8z3mqEDo}Cakk5+W;-riW>;IoX=S}b*Bm%_HKTetSsYRa~~Twl_kR!|W#Z^5!B z*74y=P92jCx$-SL*tK((nC>=4R6FT!dieI7E|0oI9cTHHqMq&9IX9Q_0}t{qW(P)^ zy09|4q}x{Y#GNL}T%IwrG^GufnicH$cVf)zkA>4q+lV_URJqim)}pKscu0Tp*a? zkCe_9xnh#*lse8kChPaDLhIqJYzkB5{0_m)Xm6b)K3xqR2wdp3Mj1@H$mV$Ufvh_1 zE~^5!UiPQ5h2QtY2L`am?rNULjY&;Q#vL;>_qO7_KDYj0^uxecx97S3`T348@!?x9 zPA^sVa2Ur%uQkb8D;O3`mrNn=Xs}O-Da!{=*2TL_k2NMW@OMgP}$j+!*MAS^X#Pa`sVeE(3FxpuglbDj(GEe$4YLm@Ey2W?yvvi zoYO-io7j5GeP{Mu-rjX3R9j=Xy=WsM#Z+0~)Q*U?M(i(jRBoD0>-XI<^K5gcyT%^t zX;_nZo_$nAZ8OKY#GgKFt80yFl$^Z*yOUcftAi7gcI){cRqY}yPS19^ob^dlGtO1z zlJTa+8Z)DHIyEiX_MuE1dXY-Q6GNFbYFZIV4ZNoI>GRC^nbAa7{$Q>oOIxY;O5w2o zaAmZd{RSLb?(33HwkCJTvqag@%w2zl*VA7{hdb15V%-heU!V5uOb?v4Pb(U4;p*R~ zw3WI4_OGM;jp@z#8VR~;Z*Gf3gqX4YMlXw3;Z(;3Oix<9etyrJ79JOuX`~+Un)Y*Q zEIqJls9-qGa>RVXG0813*s;o=V{XD-nsxiw67zGFvL%I6UV`a~8Z71O0qhnc+76c& z!3y1o&L+YKNp<$`w{8UdyHW4%c4un)4L=@tmElLKBM&#~n>3~vWk#p;x7er400t=7 zPrSSwBA++uH1+0nRm-MgA1RhCg3s!rH_uO2wN%Bcmuv#~kWU|v0p#$WiD@6bDxbcW zk0`az$(LI$sSuTm=^&a7fYw)oBS#pw(EhVptewjQ6qCbt0B)7aGTMe;n8)A1Nh7uvRzHNbm>8^l$_kIH(tU*k+QP!b7y*TL(x+eCVJxMQe?CP((sQdHB zWn_@M?v1IwDxIy>XPo0;lqnzMGRN7sI#?qvCcIrRSf6fDvdVB*&XwAp7gq;*BHBaE z0~c=Ev?&a!;k>n>HS$@!qlXFrntqE-ynlsb8q)9LM&IG0i3ra86)LwjIky)s=Fn%& zLOhP(oEB9`Gm4=rt`QjHl0DGSk{hE}*Bvldb@b8JE?#zab^?E>Al)biI49NZ&9f5< z95q7%mK-Hy;zyfXcj+4A)95$UJ$kge1*k-RnSsNr=O<<5llt>|;_NpF#w^k-0(e~p ze7mVG@1`3+%j7JTv)h|jrKh7>Z)ZVAuyQb9qM$=A%5&;5k9($k@ZE%&X@2gc>&${2k5ZG~gDoqN788(4BKWs}2TdFcu!^hHlbjuurQBQT z%DJc`CkxuG)v*#Dww({oCg}M0=_N#(`y@3_X?^U)adp~jCz^ayz_C9#%UQ0vOslzO z&|k;+#{M5dJyz~G?JyRifp{aE%YU##t@c37*_AsYI8!x(DV-S3R6)mdTC2xInRH1LO}XLMllI#2H(o<{A8p7j&Q>W&eYP3M)x@Vvx&2#$?VsG!YRLI@fkI02g`Aum zsebS0ze{ov-6f0!p}hc;$yMW9Pb{q3w#002KlNeM4t=jI%Zyw@;QoToUNro?%tV&5WO3TR!RE)Dj#o|Mlj>`@L8H zwn*Bb^WupWcA3+X>b&ynVlh?vRR>$50qN3toQ|0YX0!i9o9pP~9l0$zJ3?0QW+qBH ztOLVUoJ5a{>}(J@vD@y&)xC`tf&SJtaj`MZ#fr`WBQYG)p8F01t=Br`4SWQKy4f{l zTHXs*1Wvt9EtXsW5LDzdCZ#?z*;vxvcNoW*!>OpNQ5o5l$#ZviU#jZkwmoD;?9X?P zcRf8S;*^=@Sv(Tk)vJ}~&B?K}ALMA=l!*U9M}6AO-0RsfX8#FM+kUzFRFm9RuNT24 zd!-qS(uP$}+dj!bk+CYIzpw9lM%)SUfLkj54MsLV^}*upI)2Z;?COo5>Xb~);g`-+ zdt|NdKRIg4y@f3@ikT>$DC%0jIUSgX+g2e;>8G+XU@Y0fF=Igqziqgpj!G9FWOi#U z7WZa|e~L=cw?Xz=FTVFS*+Xl$V}760xi8z;BrHJ48-U9aL{y@Ky%+SXm|DjN28 zzN!-+N0A__M&{;C*UN0qzTRmIR=t|ILYJHuN}_DVK)|^*ItmK~=INfp?Onb11Gh9? zK>mx`WUEWxO(*wq+2z-p@7AlyCR_crDkZ>jm35;0%RjM~`@Oo&X%*)K9Jf4QWp8r( zxacCs8;76W0SLYDyq|Yhd}eoCK#gyEU7FSLg<{5WGr*crsF4;s|Lw^wGfi{x(JVJB zD^zkD1E$`Xy4UL);l2^71o&65t$WjL2c%yX^N=E7Wu3HP+Cx-_f!RiS?5Qru*S{S2x|uy)`z}ro6=5>u26o6A_hHTPU9} z%YoHn^t2Tvu#T4PV`-lHn=Ujr+Jtps?yeCDwFlCT9+(qS(>zHkv|-r8y=X(pBkwl9GUXQM}8_r+1BTnz6k-I&ezxVla1$Fn0Gk2 z`gd=PFV{CT3uoqyx7f#gSorCZbq@hULYV~RGvi;V38qH6dZ|q1A(<7i`kj{F$-bg$ zXy)WtuW~h&hs3oft{sSxll>4O>1<-0&-HuW;c8CUcOxW(2xq?+qFQnBs=gj{uSROr z)lq#_60{|Bx+rgh@4=q!61g{P`6cY@{=jecuf_}=auRfZIA~^OW?g+gDr%!m_v0Pn zRjA$@`p#_H*6=ys<6utb#%r(pGXn)PgT~g#ai3AhKKAo($5T5A_IvvrJa{{&3w6Xp z+r-!il-tU(B~vb;-j2aLTA~|e>so3;6vi_<$0>B3#SN8U4Pm!Z_T9Xs;ta@!gZje5 z=~3+sk(dJqD-G>LrITp)WfsKVT%Rs}{K`!Y@!aU>Xj7lwgea6A!rH0Bf5I_m7an@% zh5F3OdCL!P(N-SJs%p8~nC_})*Sa@2GX1Y@cS&=qihmTTkw90yIeEt@R zwO;hGf`Y*u6K{vE?>LK3NH(!2dhXxd>QJpXA0Qz0%U4bkU@2U@ClMCg2V|o<`0W#f zXwG{*|6S|&#+?l(J{5f!qmXVk4HDye`HR#ei~0eFE9>+UwY4&4G$)S7w>362bo&qc zT4Z+3U9fC#T=9$3%DPhF@;r;jZhx*9+hF-%Rr&})-{g_&?yYkdED;^+(KN(1D4VBbsf2mcIv-jN&{=r?LXG`exez0gH3T8ZVuUs}TWO(3#7O&O6 zYQ4VmscdIR;M76hYa6wZs8u5qz=~YA7dJO>YTsxMJiUFiF57-=$KacnJJgNi;t*We zW1gxVuSSNc0+um9d#gr`AC@Q@lGGIC{^xpsq+^_A(@blv#p_I)`leEkw91Sf1^~qU3jmNS+6fE)OQh;E)qy5tsY(q?up{B@*9SLm$?7M!^O{F~F9la`iN?rT!(=hqEPx%YBO^-Hs1A4Bh# z5h(0!?w?wknj;ggu~a=^szC||`NDUI$$J{p%%}+BiggAe`&uDr^QMObJNjzMmaN@( zCd7Z>pw^FUVog|TsZljPTEC|(>vh0Pr=)sR{PciLRZEI_krJ^YaX8Rgea@PvOY?=aYT?{jT5>BgNzW$D6f&y0mYhF>`Sa)@73~_7| zB?AA@9*e+*d%3DtmQA~F2l->mE&Gi?Me`AjhK^H%RLR{HQ7@FM#5rm@aBtcCn_p55 zPH1a6(y0-?BMz1Ia<@l6ELHJ3tPQGizPOqS(%(Z(*`lx{y+rvuI|Wu+A-CcbK!5#= zLhG|>u)CTWYM&L1)@p8aYQ0qmTv2mr+et3YwFROlA$zJtyUHA2*q;eB*7+_jHWwAv9-`rIk%&N<1^Wa!LJhwi^EpcLey#DhtHynn% z$-d+fVgpz%cRrT1L>{S%vWbg|^=wxJKUXc+Xc`J|8z$Jl$I!6RE)d)Z3bJxsuINC1 zZlEUH56s;|aEc?ZLjqNRueFeax*tY`nm8B9jr7#$u!*h%LsD)%d~*Ey+nr)jBn$ge zd-Jz+I1EzuoyA*NA`&Q$J<%if;0iFZ_YAMcl`5;Pw~PJxSM3|qx+b!L$iqVR=r&KB zu^VkPnyxIF(xhNd)UZ*Wi`7DM zHzV3s3<444J$9EL@&t?!brgUgP%yH_iVn6HgT)L(vbyzn5P{XC7r5CzJN|yz{_69J zgT@nh^T*Hg7q8v-XlPO)F*ESF(ryU+Y@q*9Mu>W1ot+kTjUL6eX^Q! zbcN-smbcC_d53*@j`lrUz(dKH(#4}B zK;$7HXIPfAK>$in{Fvbn$J5^7aMRyrnX**@U-0ULx@j?cC`?bg z9Fj-`=x*zd@>Tgdh*A4Ik?#uRVCaMQIg-rmo z4`MyNdz3q(z_29%<~M^6ac!|pv;wg4#%87 z>^?Vt=>{f33oG5cXoKC8Yo>0GxABlr!YEHd6xq6M{9Z%r{RL2Oq?0~Ee#-zXu)O{2 z93r}0K?^4UP=fc;B+P?*FCV%}s{(lmOVD@)1T+tIvaY#D?G;wlWOAqPPPxhc6gxY+ z4L{%cTOoo%9wLD2gKtk<@tqhVqWiS!coT71-l+RCAv0|&$g zS93E_iQr?dWYv}g3f7w5x^Yw*NcK>SwAZX4ZTq|7Loe~>7d)JSPIz2zf6z4aUMtpbeJE@$_SefHzYAt{%pu2*tsFjg06HtxBl?tgeUK$EAx~em)__W`d(J z0*^J%aQ2q7V}13~2uqsmwHC!HSR66a&l%h5v+Wy4Q5~`}5XDje$+5vXPau}b+bhnJ z-6D$fQ(^HZ#Kgwi^^DzaAf&o#!cki+Oq(0@o&A*AJ@NU^f~!6Qn81JkBz5$^Kw$mn zy+mKjV^V~-WXFQ-7Bl;R5wjl`P|WPdpKty8FVT+wc^~Z`QLz7uo?_qGuJFIS0RKr% z{U3dz`s~Y}z8uJRJ9}}f%^p3y7jg?q7bhhpt3;*as|8p!9|ML|8 zrH+(;bUEqvT+7l;3L`!Yf?gGJXbhy{9_TC&kCcf;qF%_mX9Z;>9Bv6o5w$n_NpMr0 zXx#`Zhwlnu211{;w6pMyxA5Jr2YC**Wam*~9$7La0qcg@_ETSU{tLCD75A5k9m^xLA!lg-YP0uxxa|` zeFGe-<>=_h?tEP)UITXv92Sw5a5i-HXeQ{Av`0{cNF!p)E&X>-5e-Anzi{)@Anu6p zs&Uxb8|Mz+Irizuqpc!VKn1*qHfnkckAzyH1TFu3$MUoA2-pm@=5_U0_`SYI&qmT| z=x4dWVtJ^+LYcX(E}9w%wGEul=oN!J`ZgK9s*aYEW^M2oXj*(?@0;gKN}2}pK-Mx@ zE)Un2ZHj3``9<_IX_1g?h*#CWbwK#tGIh3TPh-j9(ut?FFNj2CYQts{ad&bA0B4w4 z-Z3+tE0eO2zWf4LscKOnVV-=C!RQsLKAn*N9t6(x1a|8LPU?Yzk0P%NAh+ox*jLLa zP|4r2YC-cX<}Zi50j*b^UWegeue7by#KIyTprTvL*+(w`vb_z`;Zl4R`3eNn9fB(C zd8owyW&p6R$Qj20-_RLd(Q33?Gi;8L{6zHG?S6?gWch4GMCdu zS$>NH*RDss%zN>*FxuliD}=eXt|gohs32v6^z8Y<2+hyyAI=b4}@|q z6rbpOCzsezjW^`H_3+eo<%_gXK?h!b6}e#3!u=>?+YZF63{lxY$~rW;@ab`mEL=rY z4lR~7KM$qA4Jvi>1~C7-*rc4IogQCe$>VmB;p3zur7WiH7u8gsoLnMX zT@fyAGT)2XA0}}b$^B`msWKg>QYFB@??RndkTbh!eA9|M345iyRcGJWwoK&9hDVQ{ zy3S!-STqy=Bp}xMx8t)HZvLiRa4aJ|_uDaNztC*RTM8^Ln3ke@R#Xx&9p}BZ)QKh3 zeOO3DAiEP6-Rk5>qE-y#Fp^Je9KTNX?0b?_)q;vi3AKm@3Sn}nX+#Rmb~o-RX_bOK z=UYE5{RMj(5DG$0j>qFnPjtis>|2pl3ktV+z>JKuw~HZAVEUdn|Cu zD8rI9&+tPa^FfXR1voZJV+p5>xh9VxC)YieVBGk4l~(kV#}W?iv**rlO6gL9f+_zJ z-%b`K;>#vF0s~2?x_xwB56(vAJoPuX7dw~A?=KB%b*t0jK`pIOft1XP7x3Z-pdpZ# zYN6`rZo^^JagdQywB6?n=R86{B})7zhlm7x~9_xzLg@%~K*-$e790vd=Iy)lJ%{7c%o+tt{1Cit4;^~> z^beH!TFzO+QUKv^@x4L1c^huvmSo)Q=EoAHq8b1Ricig;&Vs&cyuD}^!y(4c&{jP% zLU@z@H8uW!iQD+U^``$PicYb`uht=M2hNPdXTmUiT1xmqFFd-w?A1C@9&mH)a4H&W z;JShp;r%J8c6Qe6xt56R%@AQ--~|EKC0D;z=e!PhkdSND z1KZpiyT&=r&b8ynE7gFE@v!ivx=<`-D8*j?u&^>hrV`Xk1QIt>=oGb_t}pm<)vldv zk^^+qc$QB2y;qs#fr5!ZQe^5^&!g|Xpm$xP)E8zjo34uRUQVtvnTn@?`F65Zz*Q58 z!|uEq<5Z#C!K)_oViaT)HsRFwUw~g}F|&cig{fCkX!oemfum4G6>yk43ilnGNzR4j zJwnvCC2?m=KutH20^rFfd87wG$ z7JT;4LUgV>Hc{zQYAnhcnS;OYOgW*hGzlFd!UFq-hXlaMG^niEN^1lNtdRDr`*i|>6^Z7 z^);FNUng)x$(eX^1O~Y{V%u4?Hl3h*s~((PNy-OlXaatTAf`XPL`_6}>M*;L6n zwgs>c zPId?uwG(6|PXh+VGU*MjzjtUS@Q}Dp%Qjsz8|07mM>45<;O~1wryBMS7Fz}8kU={^ zXQ<99mIq`jyD&(1C==G`OFV5mJd%x$i7Rj-RpmgBsf;WhO<-j-J1t7BNB&qC54^SD%=Ol z50}oONRx5s@g*oiE6Lg>>xxRp1}HW8{RLq1sy&NCVYjon^EasIP>*Fv49xTZ`DB4QlXmrY+R#7H zc|AEz1NwBbBZ*}<50-Vq+(=20%JRhc041s@UF-D?Q5EPX$kS2>$>?WGdnu3w4(d_- z>lbB)zWQ2W8VRNz%~p=>ovC}Jy8e((djLtp6=EJy=qxtXj)H^3?UxIs`lo*>IZ^ra z@Y$0=L07N5tl+#}xTbyOmgI`}eJQ#SV*>i1L(?ll9)IxApeX zi0D6GGq-(XF0x^BMd>&5H;(Y{%DFPFKl?@Y?o4%m_G0_n0dYkWJ<6IBnG?Qo!f2D`;(Sup#)~|AM$w}c2&2@uA(+zpP@V2AJ2DNOl?r$bWMp0xLVv$Wb=AN)5 zaR}2;JF38f7y+ZyjqN|!qH=8D`F*;0?E_Lup(sniZ~hQQ6O+LS;n6(3`T1}9E!VGK z7pgrdu%pxi17Li~s&*IN>=Ye|nA=oT!=z#j36umNvMy=KH+T9AQj+iAzwbBf+ffO_ zO&k={pR;9I@Fg0sEt6Ao?6Uo-lYkA!jdN_jFaYiA(AaP^_vX#@6z5)qsGI+BH@%_F zvm@4fq`Ql@xTiYSkmCuiLKx4pThtCeEoHLY-Nz@?rJn z7pEM+Ec7;{a_nrLq$QI(jEwfDAyE6h+$!$VF@vIqx>k$@>ru?AvbcEGwDGSYU%?Ma z_HyVA?YWB?heT9;Mh^{tabc_Xu)mGIJ`AcK&1XD1>z-v3ZdDnPI#w+4u@@0HxX`GV zR3HW?rKGIAXKO&Nw3F-dSCjO-zHIqO@pMPQio+%sT}Kvp;mpRA85MdRae@LV&!KFA z)t*S5$7nga{*#uPNw0|y8C&xD4%J8@@z@tRvp#4?lMa(bPN{=O|5m_qfd!<85;iGq zg~PS;?d{<1V6owGON)H**DihR6=PGjP*gtUs}mOY&MUgQpiJ-j^GltFi%lGEZs+ga zyO%^d09F>=q95F#gR`$#lms}gNCax7d>%7Mb^*kfhHI-BO^;6Px$(2UzJB(u75gBX zDpfr}bhi;3G)nNk)Z2nh@C9ksgQ8RGdUA4Rl{_cx`B&23uxf=CDxaKaFmX0ZO66VU zA*yR6YR7JX3_aCfGGouStdz;6#Msx@XT0ElRrxVJY1gk;phVBpe{odRauf__1RzH!GHHgCo-(`+9nAo&r*)CP{^CJHMb2!a=Jkj>NO{!$xi7^s1Fzkj}%(k61^Vhc6^oSj;(=yJ3{n4NB1 z)%fDoVwPZLdLsxb(i^>zYq&T@mNAYwU4dI^)cO)+Gzq6p|{yO^OmOLr%QWHJBPzo-%9-3BG zR!Z)DVYnbfEy69L$(`&^kIbj;FGf(t-rAsv2ooo|utd1_P$tG{W_N<#Er&ah(zZ}V z1?tf>q9#zr79*+hWrZwizUy+0=bKiRh>!^bo|$Hc%qs4EQRhyW3MxV9hBArg%P3zJ z(@EGUsY1B2tCg{39X|c3WFcg+(YMtJY+Qcfhoc6)3g`t~?)q>&N!YG^Nr~NXS!tI9 zOl#~BhN>;iYs$Y^pwQvxig>&MaSVK3;32tIEq0Yche`=DV`ja9iG;UjgRq&CVSg2; znqnc7dOxrR2AtfCj15Bc-s;)mpYRT=uMKBDCF3!>ut4Q*bQ7{hC=-SO>1}UOj8zhI zmb0E-o$+8}Q`49-%l0w2Q@i@+==)4g1u0G1+2);JEH?pt88I|9480tlJKMkA&y zz=nxqM^|NJ53;xxwZ!Cn^%j_gC9pp%jOVxZ)WTI%I5XMLLmN}voKF{LwK#*05|%cI z<`OJw>x#Ve7LaE=N%(3W%nZ%}=&R`_D+N}3Bi6)k;GlRVS+)>C){?;%;F${=J9}W` zk>`KTU%vUq`)|wQqF--)tg>fy$uy@#g0%uYApkckQ^&~x;EF|wxGrp`#aUrXEQ^kX zMJp*Str|RR98rSd;iwZTDk`YQlmI?;bDno^FiKb}JpLEfhC9r4reL;~)c=wC^Z+?l z?UN^AxRko_LS5^^K)BmnXZz^8ZAb}@FZG1q7Z%Q*TL^f>yDbuMh6IJ=EdOVpT0A(R zHhR$%&OEe+$yzO?8O}3qH;rne`@JQe1P;$Pz+UwAZ+O9 z?|qUWh=2(c07$f8Dn1RGZO}VHHh;)|#%#~KPtlwJW$xnpOQq1}rbnHlp#FQPTN4nd zL?fO&fI{Xww`5udt3m8NnS0+q1nn-BGAuUWk`3;>kz)ri-#XzTG(}DFZw?{Q9-E0w zZhU#>Erp5;x%%(N&{Eb&la2N%xjqQ^xlGDCYbxKKMroWza;Y-Q5r!MD1&*BRfM+84 z{pTlPgiTO3DLEM@5iMVSp!0RuFx$qj`TKj|#vtK_qIE4HKVR5_I|AEFc71?rPj%I` zJ(oAYGvzkasz?USHScb;rkTc(My9V7j zr$r5Mp0Fp$D?|;r_3CmLtrWg)#dcuOb_5=3*3i6FfEpyVwg=Wa4R$MR*@|%0XzB|e z2CoZHhKj5Qz2c}!eEO1-Qd6twCg@L2Q==D4GB%7!9o{1dC*gkug}v;c2}IysI2lwT znId76B#=BUB`r+}h>`%S-7G_RXoBSQ@a3qccd`B80oK~K?VcUqVDH{e_@i{#1~$C6 z6>dJF6LB!1On92^E*ozbh50h5K)D7 zgf+>IhNe*_%jwp$aP**w^fqn8WA;+1%>cjlA+y|AS4;ffp4;LQ6ABcW8ban4Gl~ z2$FP!mENO^VGh3wvoJ#=x$T2?1hsDT|7*h4(~|r)rT-{xTUr3NtQ8rSWDVmSIeB^O z)>}vDp*=u>P-_#o0|%rdgNk=|NgH>GN*yWsf~ zfp+)4OIoow7KqS_Xs&~)E1tRZ`~6$ko;Je)h`}1{DROEO1G6i7(;Aiwx(*3)y}!{# z@DA!48b;!7TcIh4d|IJQ5)@-wT3VRp22pjXW986GZ&?jM*@7W=itt%iVkOk7_{f8;QMxnrfB{s{!7eLBnCIEht z)F%QojzZ=Gr}qAnOg~9fLk|qnWo=W_IRtbfRd%xJK-@jPz`hrXWh7cMIV?6g?f=bp zfrR#R`ZJV$cZxknkM!UWF{-^GeJv`_CH zHT|oqmJ_v*D$sNx!=h6TzjK!BHU>jf62(?^i|{%BpY+%LC-i`S z!h!e~&4c-Oj^)4fDXwe!F1b+*aA;2=kU!1>RG2#>OgJdAQmkLWaC@1&{2S34w%5SG8vG`#6K{)fSqkt5M0)kegfwA!*Qb zF05*PJ28PQHhOY0B~U=39yzs4?z)xyaED-KA|hvrh2Q#()(;<`(ZTSOBhu2+L}*H~ z(n@B0GKJE4P_SnQo{7VPrYK* zQ?ppal1}WE4x23aG8_L&cquzjWm(#gTSIZX9w6^wDpY6Zk*2$y|XID zp;j21rq=z|*=BVc@$IwR7Yv5@K14edRTlmJ%72P=)d26&2nV8SCdUUk8AUiEtd_!? zLpU1jPE^|tBTe}1War}0Q*btdCCFYQ%FsxIMlxvjf!n^SiD(8R<@#mS+oUVhyYya*=_ET52K&hy9NjZ`-ylnJ+)Dq@R7^ZS|+oT>aD$+f1V==&(%lCu$ZM znw~|WDdKhtpP$L1)NLJ4MF*DGDYNMMbtyD^@V4~K30iRP>p7wZAZm?QeNLx1I5>C^ z-ssZ%NqCq|J!kAMubthFfIfu|q;fd@&y(8%y!E~=>BEZ_Eh4oJsZbYAb-Oo|ikW2I ziRo#f(gX!YPO~{m475xDkaFyt^I-WAcmQVUR1Ka`1C0}?C@RVvi@ z|FdW;J0Y0dx5VnpS2N&J=pCtbpcA^IiPqP=bdbx(;ZIH?*4Z_#dJtIWAVf@sT*mG4Bh z|G@aK{47sIDo2vNS_(d7s0AS&<#>L~c=YVtZQ-K)$3}K+r@zNkt`wXkwhIFCC$<&tH7P=-1pKs)r&ir-A>8mu3pJRG~M>H*yvuwGT6B zd?OiKmab1@+$eU-T5kpGpM6n!Q0|`~sGhXL<`P3CA&zCmZ%5_?jomw!(VZeDyj=ft zD*v;?{r=~M{^QPc3YMZrktMsyW&=3J1)bNmQk=JQq59M-Awarz?a(k+ctnlspUrLu z32+|fTf%cFN7<8U@rTr!lo18t^Kj*#!j$x0j}s$l1rwe(etcEb}3y=*Y!gV}7CoNDaYfrjc(=16xH zXTO~Aw4@|0ESfuv>njxFq2j{d%3?z;k`KA*(3}x*;g%eTMQ_N!IScha*wPE9??vZ1 zjD0nyq!}r)Vvf6P9UM#n%^Gz2;vw@hZTafw5H=)P?S3Ow@*%HDbOBM~lnEr}Tmv<> zSoks^6cURx85d4{ZOGVXq119N1H@2vd+yzJk6*PfNz~uFH>)L4@l9%addD|!AJp$N zF`v1^e50EOwz03EEyZ!17&c<4rp}ctbLL& zI@m2W#ImZKnth{z%lsY~<~2qe*bv`Ioy zo2b7yyZ!%&nT&y}T9SSgeszxiCiho{69$HcJ*PHS?}UO3eM7_wll#cvPrL!iW9Xpm zR+>|YxpMtCjrVU&zIN?_b&O*~XPd=LA+$`ww%f+rBU9joGcnJYTWah(tl5BWSut6o zqZ=Nf(KEXxz0xP`^5Mye;doZ)sxH)B!s(YTGZ`NOD7M^l?{u}j=jyf*!T`JZTQE-z(05D>c111_ilqypy#_}Pc`&) z4ss*z_bJ?gp9*K+YZR^$sHYptaAa#hk;O2V!VEzc1CO+ig1##N!8U>+RqI&Poo@v! zf2ijV#_H=Z7=`e-ktOGuiYS3cf}Q;?$4YXY@s~^M_Uv%&w+V{glwUv_Ub5OBDCHx+OyDPQGi1@$UDtas5Qx&5)sNN$@ zup8RFZR9dWMLmq|p{O!w+R9D@QTCW?n^}D05X7Zch+a`j?sg=ZkuekV9ME_-xViNH zb?@b0LjWNcrn^!DsmxvBhTV`cs<7?_wXYk?HfN^Q{6tnIGNTww48x(DgINYN(o+!{ z1WC$%7g}aue~^Zb_JCYvB=fNP)kXFX=3V}<;zyX}k~vG@_J7x|L3TnH(5Br`X?2Lb zNKHrFf>JoR=SA(G8c*YbbN~^fxaEoO&T{gD!FsPe{^5#QMl{mb^>Vi!C@q ztkc=W@t-$zMP18_XrSS^n14Vnh#?GUnSyq!jkboyTWR(F<72Ng=X_YWWhq5bXbU;v zLN*@SmN&ALOS4sE7f4hMkjbt{GP!!;42O@%vxFE_JihMoJp}MjAtMWib9;3?;KwM{JdB0dfcQ!VJd30nKAP=2x67^Cs+ciXcDKB7OgIPf z@+i1Z8f>;o`#e}({RB1iMg217Oe% zOHL&ktAsJ2;oY+t_V?zSU%17X7%Qgd^gf2`y^io5tFtPjK<>7Yc>89@N zXU;ef^=XC$`Ep+BFI~%{jYDlN<&d^YyNkh_XHXhDKtZ(6vaq**^+A8+(!wn#XmZ+G z7VWcsLq3KiBop50hA_&4Pa*>wdKH|5-ql^@bThHxX-`VP-OvmtQl}(@9ztpuhYoBb z98(EtDmj%32p$H@#UXi;1@ewrP`ww(z7ZPEi)kX_T(AS1q3b=M0Vd>NP91l4G`s6w z;OIPyFtvtUn7FZ-_`uY% zBOJ7gX%7iGIph9;PK>V3>4p$^8Q;{4*_g#^s$HvB{W4Yi;K745A_iO^6zToJ6PfMn z!s0A)XN%DgUtP#2xJ;zsK5Uqz{u;7fQWBscD-pRjB!_2?&yeGXjDPfjyXOH47I4DUyy0*{k1v}hj9x{l4H`O-*cEkvk8K~ zKRZi>kzUpGWB{{7gyP-7(~7Hri_i!0fMgeHP9&c$4K*NJlorfmWUmuU2xcs3;u##i zvNWyGj)qlaW({9faI$>+)Y-g2MCj(UQ`yE-X*2%nsDb9^3UK z0*eNQZ*$3;d$Jk8hkm`97GyeX>}%a(3o68IH6DiA&e)dNDA1Ld6@z@xi8;5%y{n{H z3sqi~l8X?OrIXRt^(=ZMW;0C%?hAx9O6YFG#wC}04py!L%+m->=fVeZ^nNq%f_n)j zE`}D0o|ZpRAZ%eg52cO_4wqk~47w|?8O$aVzVpYPp#l^2BaZj4TGM!7{L!<-fKP*| zY5v6UavUd_{7)wHZS=yVQ{f)ddG=-do+@y@<{Zfn3abnsdO8Y>g9cuQ# z09I@2r&Mx#DK#xECA9&~a7?4#+kprTnrs?A4&QIN#x!zaA9XC4c1g@30uFtgb-Uu# zR55A9i{55zJdaVKKQho5xX?N@Yi;9=RSn(wv0*bU)9p($hqL7m>SZ=qd&o_6fD|*d z?T1&HoiggmVxyf*;Lw~XhdCYE+@ZCWj0SYk@zsB8i2FTS4k&$!*E2#&G7kf0i?w^^ zHmGZEVu)2!f2U=5$8QSuhYsz;HN%{nGsu)Z7}Y__qCu}G(peAEvamUn8iY+8v-Js0 zP#xBFzu)=H*a7^Afvxm20^+BEpB$oW$H_&SsPp0OVoZdIh8f7JG-y_8fLPY0OW)Bn zl{mzu3Dt%c)7;Qnivqa7MkWR`uhJ(|({c4GYPOEaZXU1cNl8o7W|K!KwRo2!1O%2r zEbr6OeEx;T8Ien=HkW8XCT_8XEo^dn2zOjKgu37qdPtXrItZKLMQzlSDb4>9Tcv?7 zYDa2g3KUF|nql*GrAaXW%+ij-;E7SC(15w{91*ly(m2l{e863CDvC6qBml}s55_}! zW3KyoN?|l~%djO8bJ6df5UIgD-t!n>C{=^PQ^Uf0&VH+^#`gMDktVqL<&^&VoSeyH z2f4S)uF5G|+yCrj-|jmAFKbGUv+HiC2sLSD9D6~uaoff{_6L7Lo}<|{TNf%aSwgcn z-Ru#V-2-F$7!BgHWFKAJl)kSIfkae zKa-fla6Gh73jUcvd6JdffLNawKlW(AEN4jLKe=3Wgd2Xkp67CHk8e_HVJuyH|g zv7yNiPVlH2$Ge)PP1|3`*2&#OxL?JB^Ytv~d%2l}Z+9Ed=zORY1}o3^EE>TRF|Qus zhz2FH*fgkA_yYz&p_4%qABt7WT~wdrbS1j+t`g08fxErypg13`syM4=Ngd>?hoWsm zb70^J(*^V(7(5-gJ7~imef@AiV;YBF@BTrkr^QDjKcx1$E8E7~PfUIt)1Z$&nmP|n5M7au5%9+K85h&0Q<0mJCg$cR zEDH=0(JuYe?Q7@4*V~6+8eo>&z&0B(h^28fGC6{AHmO66Q?*guG3SU&>405ggeIeg z&mhK9M@y(>wne-Wgt$8FG9hpo1v^GcAI+(+^?5T1o`A;E5Br`PJJO~t-qdo)b!>0*=Od1Z3tI%4Th;{6;zgDHN zwhDJkkn0(E(I`S7UfQ8|zVfGKbyg9}&IbOmDe25iRK+f1w*$W{U9{-uGfDH<^NfBz zHmb9Bcje^+Qr}5jJ9+xYA0jtiN&0D?M)Xp*3{NqDTxuB!9E!m&@I&uWto9j>bp1Ld?2~6-U;0ijP-tY%+RgnfKeo^NC2&XSr>45b#>Ps^C(zwZeL7S)pjcDDnV&A}2d6-0 zbL<)fP*q-UV!V(jKxIfjbOK^!IBMd7*1XK3`{YYUWC4eTWnCE`IO9$86I%hnyr2Fu zj|Q^Gc@|RB4h>8I56}Y(a}|2{^$ZNgJ4!wZ(tJoit_0A;5>XjgN0Y|hLwKjDTrg{nH zP|0ZT-rvxLpTJUqgSHXyvlSfxgR+a1>k!#N!k+U(>wI&zy&?Me$SrpT+A9?7A@06U z)=zkjMP1A6N9hy`v6g)Qpsg^W{E7Ndo5@T5H;12Tv`lpoC-G=Ojw2@meNM{Q8T$@H z@zra6h280oWZ=PJPi3>ee!1eU#JV8%sLSR-k}=)V8&? zMZZG8%@ZF+u6;{<%o&40xf?&Bgq)Uy`QZpP9r*UPMCCh+=O1CJZ#&@s_)E*$`K}lD z-}#FRJ0Wvk&C1$@Yuk)ZaBl^=FPHp6qj{>I0RdD2m%H6O@i?iwyL*#@LSky_-~o|2 zjG&WSNUM!~1FoAuo*&$FOMF)|7_UNz6npeL{18{Ov&RmIKmk|!t5II}o#o{_`ir2A zs7>gd(R{R%Jf*bUwn3t&`KxqlVVj^LWmz>}>Z#0_e1KAZSB@P++Hfh!@BN4tp4@tUygp z_+bXsJnv)IW-+$ZFoh4aD^^Kvp2&)a2ri*avXCuYp>hrr`>ZkKVQ?BP*RcdMQd3h2 z|I||pmf~XM$a+oeA2ROOu zck^&QZ9tvdIyzR75v?jlAu%N-Wh*iEbjA7OgYjn z0G@v6%bXI)q(NT!`>9vl(NT+?Vo@@avR|Hs^O^y0$Yj~uqoc&`6Z6@Opg1ksHM1KO zm&<9;iFA5lMqL2;yoKgf7>nnm(|W9;^;n0wayo4MES=Yd{*qeJVNBFutG)1^KJs)d zr!;BZ?d&wD*YRcHbY*2_C{sp8hR0v@HLWUxcKc1FhK5GVMCCM1I^hVKZXRUOa1Vse zEBobOI2Jz7_ri7++~68cE=I)Ol{<#PbnhNw>aQ9@M^=f3`!dTBRy`W>Q2Jy|_u$U=0FtUC>$tkQ;<7&um_oP8SMPn# zdU!pUfrDpFz;l2U>BQvHoouv2(Et+GjPWnZuzdTW+EhIJbPF+hDJHq=z}_6P@Frs_ zO?VXRTXKn>!Gsku*L1E%m-Jz%JmZDOewfJHPmZrCmu zz>j`(M!ALWsP?(uV?DTFn~Zk@-gX+j6Y#IRxocmGB zmcih^LjTS4Y5Qo%V0^=1{qX$>r_kO;XFm=5X@$X#9m}3S{Pl~imxO|2cZeJ_i#CnQ zD?Czh;!E2lw%Y;=btBdC%+?r2-SRgLy1&!NrTnXkANM>u*s!an!RGswKOFs=(fHKN zyFn?jXUm@NYH_-f-?FhJgKOK&bvKA8ZJ0doRx+7ueK7XK_Rs0Uyq?AOcF(6jF-!#K ziPG=;zhB}skN&V!@#z=z`_ZHG>2?@iKYuyMd+AqxUTf(Kk&&Ne)9;MWH^1Y(+t#iB zg}2}=@K=m#a2?3W7<~4{@NFMAW{6d$smoYzkg`NAbFw!@RpiOLpJhM2yQrsU#2FeY z5;e{ltuFGKw;vrVMf^E;kE_aw;}?BR zj@wXC)n@nPV@rNub90zuQ*Bin)7iE9GhF#Wp?7_Zn>TM}vh*+oiSouaXUmhRT&h+p2Y}i$8sANtqt2H_qpGd_B!=Zmu%S>)WCE=$n`b50{Yo zK*cDxA>WAC&n|so`|-uv7{`I8o#}n`9P94x?mg$;JyoBS?SIme{QmL=qYr;v4cdL? zP>A{6JL}i4SJTi4%*@R6Q;bR!iP6)}vTE`k80gbnh0FUY$jfnO&QPInt$mH2n9=Jy z*(2pjEcR%PfdVc`{<|fUSoxNi~8q$3Zb?oQd_LEKOSQHU2JsB z9S&=g@SL4&w(X9Mk*H6~%#_y9J%5mWc6OR8I>2Ww*_UI5tMlqV_%v%xdvQphC8tqh z_EWwBdj}ui>_knT(ah9%1}@e1!__VADG}2aD)Qs4Q#0JW(%%kVw|TZ?iyv2E?CqK1 z9)sD;xI)i*r6errs@5cr`UX9MYFnDnEI)^1(6Z@wPzQYAxf0 zwycXKZ^C=%cy@YhSM0xS88|U9abvysL)n=A{;Yv;9qSzNw6wIiMs9CPX^B)vs4p`% zKK^w|seH)q7Z*92W6e7ARJ-Z_~+>CyU2sYw~9N z-`qRAg%$4H9+cg}KcLyeEvPSLZ3**J63Jk)Y`Be%wB>p~T&qizt67daD4YKJ_j;A) z>3+we_x-oNJy5r?U#LJbZ@l=%*So(f7#ep~m1^@-iA#}jnfjP7W?1OQEI7DUFIDYH z2y?(GU&f|rC#N-cvcD)Xt3Mc9L5S6#H#<{VRn_*(DwS77qIsb>DoJ1AJxgQ;LKD*K zwzF*TARRtrQC#pETQ!*e{KJduVn03mWoQB$kUpMu@Zdp@iHeAT8qdl4W~2Uw41Vo1 zj$=2i-8Qkxvm6I{@p%d9{JT&8V3pOEZjt@*Wm`@3ksx+~?-tI))3j^XuDMP37^t$R zKKgjRu25BIoaes)AGSaLOMFA{BTYGTNyO6eqevXUby_*t+Gm`@mg4muxlGgfhlRo@15`RX9O+K>aX7!tu z0{rT&k_&|5EQZsP+!Gq2b93ccdmQ^4+vgqF0cDr-zDk*L@4zM6Joan?o6;?jpxyRP?%1 zl&WWB6{W{?&0GX_>SgX_?QtFJ4Yez0)w&HASHx)~S)>$8NR7Ap+BUm$6VlVY2j8CY zX}j;MQkc~=nmE#3CA{gd_v>}Wp|@*OjFXb5ZA;}eJou(3xuUbnbdDo%VP&GDqlGRm z;!`Ucn!HzU5_G^Xv>) zak3eUFkT**>v#9Sm1vD5eR{A$NABFhil%DL23J-#4Fm+l_oZ>3(dyBepyVJ?XL66O zFT^_6S$;UEwDqvU#Id*mL2Mdn-K*m1von){Y(KVN$-b5wDhf>zGb1tOPcJMe%D3oh zMkLXD7#F8wD0oG{#U!HH41+I5zdV+Hccu}d^OEI)pDQ9QA4qVrx+eYpErZTZU$ z*4b(buKgJuiuwJ&4nH{2{nj?N%Q+Rt1oW{TKK28h2p zax=d#63i}a8Az>AK_th?(r|4+KnkcgaC(2q{KRmut|Vox@iaV+q{^C*q?RSCgi`AU zD2S*598^_R`LB@aeWYIBgpCpx@2PBVR_a*v(C&{kjqU#GuHK2C94ESB0=s*9G;|0I z#3yt=I;0q-W|ceMJ~V+dULBhsvhVz9CM99l&=sO!CT_&IU z{pTA3a?-Z#uL~A#u(-vnHH%BI8LiSVM$#Gour}|`+40mb6eep+3*GaaJsJgP2CuZW z^>LbWt1<X!#`lC1bjhmLJDU zJ>@m@>`$^=%MS1rUs0HPzXLJic=VBNr9ZhN2cC@yb13BI%}yT)k=VIl(TdmDsDb#b z#_<8q*$|emkpJNeo*!04ylTj>6~n<1Pc*zV=E2FN?cyhj2-?j7Z6-C^kcHF%a8c!zsvx@R%yYBne z*}XoQ(c&#rIDO*eNq;L~PXD^>iSD>4;7Hr3+g=Gb-&uBvSw7rUdQUBO_|Txxw$aWv z_v|M7I7yZL*hZqu1!SKb@K~cP&i{MJ{mLYRymGAIfPbCb$3Ip@6-(5|H}uD+l$fQv z7qB-+6&4l_jUBwcCe7(}bjL@Ltlskex_X<&_yUvK)iocnt|4ir78X0g5VD2C0=I;o zd!|@Z#@?)?`1}f=C?`nP+2#r_K&r*IB<4gbw0KNUB#cNU?6Ry&HF8gN5TP7;c-NwV z_1dW>$5KXt10NmtKEB=G=GC`X{A#wShKH_FmEG;{FFgC|A26nE6i>Dd*K8eAS;dy= zf3oSol@)qjqcb)Sa1B*VfkdWteNS_4#L%RVkEe7z{o?tJ`3>H zyFc7HR&QOZk04RrL+>(!jT%rRn)~TP_S8@jzsD5eS9)??qdU{zen-f&uxG9Ax$oOc zBMkJD5)z6#6lTYVCV&b7{qaGPiAxm~80KmRoe)PM|MqAh#AERHs4y?5kU$K+|obIM96j zE>pnwvZ!bDuGdS}Tciy6EoK!~9Ev{&NEZ+oqOxevNI1Im!NZ4z$e?e=0_xwdkpJ{< zO;QWh3^}e`k)#M@qn#=X-i-Ks7w5XTxjVkkQR%S0Yf->XVbR&qBu~Az1Isi%z47tz zL$caJn7!$vw`1zt6_ksA_HtaFQV_Du^$)*0K5Y{z_or>{N$>W9LU7xv{rZ>FdOdjyR7ap<1L{ej0q=X4{5` zSuk9xkXJr~S7?SieYnajz~OCvKfE~UX4TDW&1Zd;3yYL5R!Q~%usCO!ZmknApKL6@ zUDE_qrlTJ&ZErT#H1nxeak%DY+2A}+2#+rel5B0e!)Ozs>fhk4^g z>oXq6FV|gF02^SXSW22e*aMjJEha64@FN^?+vwp0nGxKYi`rMuA zN65R|yj*AOR}-Tg5`cx<tPhky%8jTsi~~b+oru?Bc@Z+mn*M zzFG4o++(6^$VTV+>$o^lNh~Vl zbu80PhSZIASH)HJ9hCN=imoc%RQj~-`)8Mm4_Q>#)WjqvCRTYS+!V0hnW}rYOVEjO zseig>yI90FQPt3WkG0-u8W==ShAS9)HzYUUTuS%nGHmj;+V1SH^}%U5B{CCrS?_-S zLOs}LpO*aTN#LO(EF2-D^S@E>kt5+afcbySb2su$~yMdZXe9gSSsA} z&K<*Zv%xNDf3eo!wU6gi3(*n6X|Zu}MFcRj0aS|LK0b5Yz#y+WAB>AvHC;5lM$`bX zqHo0tDjSXj?b&=~qs@;(Dxq&J>SYj+hla35Z=+|LWXqGI-h{=4EZ`<&ZuUfmy1Z<7@kX+Po-BGP?tmg_e zdp>vYlIYp$cr9nZH8G^9Q2S~vVdZVRQKE)=&WxL%5X0_@-!{)$hH9TSoht?ai#THL zHz%DC14A00OseGJI8k<`uE9}97S%?a<#6XcpMXF>)SN;I!B0;YQBBJ%hzisXzI^%e4y|M+4~Vrn zGExntmsU6FobMYQXeRJ(T%s^LB|BPq+<9ia!}jUCMSdL50oUf9%G({d*KgigUZH4? z%5O)K;Q2pRYGvA)hHypw6$zO-yHk=g-FvEHup1@!;c8*Arr&}RmglPKaOv(nrw?M9sa?3yxC zzhhVypVZJhi6_=*6AEdc{P0lK#Y(>F?5<-mpIjT{jo~*HP{LE4ngNnv8wJ5 zi|;mo>=2E#-$(zKrRS3XS$claa%!NSUk3j zJyEze@6(I1)mcmbe8%pm>dhqhUxwf@&lVy7vv)*@I=s;3@VB96HMc+e3A}*XiaX|J zX35+pSI*n9J7>JNi6TCYjsvrXFsXzE+r(%0V~+VAwu@dBk^+8&9f6z!q%z}6e@GDE zbp6_hXjmn`Y$cv_^<^(k7>Z8Yjxfh)>?0N%$YwL~LUglhtnG4d~q9WmW&+!pb>M&3l# z3fDVYI!Ph+T7VSFvu}4JZ#CFeD92gUl$G5A-pbAP0DX`b#disW?WxA&ITJ#uXFJR8 zI={QHXb(XB4&5iad}N1y#D{VX2jBfpx&F|NuWy6>X$vwAF93PeU{PtsK7xc%bw$1} zqa$3ZCqBjI_s`cdwFGvBV(&?Ti%c1<)-i6%b{>B&I8(Nxw*i4HxbbD~v%&Yz#qcRM zPd;AYbNdD;-(bbQcU~h%eihNmMWUYLMLTrTchDiz#|Bxke7(-`W5;e2ELD{sKg(`C z9x-u((g=z;Bk+`};K|r$pL)%!MU6c4?jm11XrL0XX+^eS4gn~G{R1_`^}g5XL$;*L z8m@?#*$aSCGKu?h>pUP>`0bUA?0E|o@(4TxSvHCX!|BnQ(w{&7Li)EUCci5$kd5{? z$`G=VXtb-?h`2J|c~~JBjAsKLP;s<1Zec<2=O+AyHF}Q4H`a*}_r1}54-G}N^}9|z-99-% zq)n6Y%Ny(B-1(G!jv-5ykHsbBy7-U24wK<%Wl_P(nI^#6cfO`tqTb@TqT0iU54VFD zvrCvTqcY2}$vMGe1}Is%3lxj>SRJy~oA`+6jygc=n)VjQO6L>X?He*GM`R~xd8Cn{ zwfk(J{jwwqc|ansql%cU%3fU=z;@6uM5Rjz&VsPFP9MN&m5%RUt?t+u3bX|NQ#B&D ze8${{pKpUie1pU9KYzY6UFXTUpE4RB{q$VDuIaY8zhd01)i!_W{yKA9o%?V|vqc4W zirAIz4w5hUU%21nk3?CWfx0LK@R5iW${Y?&&K)B?HKP9h{_A#~+Aohnpaa3J2&IV_ zqB-&2#nYbM$0N4;ByYrlX+qT&ot(S};nss-t*%YMW@o4v!-$c2vW7n>+2akvG7vy) zT5lgV!t;(r!GO!OfkMFB!jKi3e&XCpy83*D%nopm#rg7Mk8ue}Q!W*YLN^y++p6hX zO?-2&WbngvQTh=THmcbr#a36XeL(--5LA#aRhV@vL77c_31|!!5`Qvk%QpPAXaS$> zEeJ#RYdsRCkJS8dIP5U1BUFh$Jd%~fj7vhNnrEjC zVjBCKfcbRot4;<<*&5M-ys%&q#o={wE_RfnUk8b8p==xC(~Lc5w8?$TkvnDY#W|yy zS~d2SX&N}x&;n#cb4JvNopPIM5iCV=Q`6E4YRYjnhA2^lI?*3hm<%!C$tdM={H-D4 zc-fyg7+d-5m&bz=qd}!^LP*gAYG4*>3_Nr6_~gy6g8~l}%p7U;RW%wemK^YF$*wJr zQaY}q6LQwMlM;8{=KQ{;2IGYqHg68D-OK_3d0Wq+HtM=?x8|9A?_x0UbQJ&~#n025 z+N-0?!KHZhKVPiBv_q$PLLDTV7?}B)%5C{u{T1n)`Z5;@IK9)pltt;5V&xehx%3x{ z#Kgt1UPD8)zxrFs)7FUSY6=*7yxyI^*fGnC_pbIAyaJ!qf@QZw^c7;2ps2Qc`OOG2 zVxXYk-t!;+Nb5F7aHH5vgqY#$JG)IyOia?IuU*-+f6@n;Mb^W45cO=@gP?C50=1aHY04BX(lO2$!@<#9d*@={ zvx$DkypReNnVt1kS(d<1Vc3T&y12mquc+7um(qqPpo*HbCbM))xYKu3v#j5+!Iz1` zr;zAEZgz(pZt&GbKKVO@XdvMj7fzDm18gx_Gx65ixY2M$6Jp>R!Do`--Yu|Etj`~e ze^&>X6q&VZP)!pYSS_-R4Vb-l+UfHaZG2SiF0ps-S?DyRg@obq5@aWmHc=d{#eX5i z!x`HTTwYc{toAZN`RwQ3(+*X7^P89?7p&Q-UD=yP4B2=?hkVV!)sKIk+Yzt7TK<)~ zp`jtMjFpv@t%&NcK_yjZmQp%)1|4Y&Y0l0BG7+)omqFQ~=; zWx=^hPjCNi!Uj=lv2jHV>HVeQauo;-Ot5ZBtNA9u1+E#YQrLCbdCN zcO|2d1t#`kTwuNw@-C?wYc)Qz`vq_!3Rp^Sa;a0zes}jAQI$?1utvvmlFyE(=@}R_ z&%Gi3KD(g;&;`$2Byq7<|9war#WMykPp$40fVy4V@2>m?>+&CU^%BP(lsA2ik^1J4o}K{R=>98R>YJPb zcj^TEgFdHf%IJ?@S9U=cq20LA_VsRO&^B3_Ji~<7ii(6ogZh+%fIfyE`?&9Dd3jR_ zdMUdV-1jT+%HVf9&KCM@YpX%*U@)%5*WY;OPGT>VQi9Io&(`La?f8RV-pxsVUx6lbsROz-@l$op#3 z`bMMA1`sO)P;FLLSGN~ziFi%toW$R*syGe#PrviE1D7izVq|3=p&#;EaU2mzQBUuW zIy1xxob6knfOpWrC6|CPX_^5#RzZga-S-Z+7b{_xl5Frzh-JFg_376U3JS;*?$_(N z_11Y5X2qK353Y$rZnPgA$~&KfEem=W(lDb_1nLqmX^aoVDg}!4(brV4;j}#Zi7+l- zewzOqmc6~bnVDH#nl)#!Q=o`mFyhv})vH&NP2i7DV7@8Vc^^jr0EjmQ=1p7 zHTJqPbdLMTPnZB)ReU&a_2m6!^#!MKdrbmZ3OY?yhK#Zm!ZVt=B48_&d<5j^GlG8m zR!gAWn_njVPlcF=LjW&|aVw-a8j`O4Y rv>Fjl6W2L*D>PnQc+weoYdRBXK9qq zXW7aXeT@iw{+99+Z+wgH$?ImGZg?+;vrK(@rcJj`8;~~xxCoawp zvjB{BYGEJXlQJ7|eQIYMZ@Wf{*@JPr!td}36lN%Pf*id9GToebhJD-~HsEHAEa_IQqb<;kd;c5YGb*Je)z>=Ol_yU5oX#3t5>hOsRns zZXoiSswnyf!|tJ#_|Lo$bV~2L{|SfC|Gwz;U-2pZ`zrohZD0Rh&j0*!o^-UIcaY@o z6DLmC*{|Eb2rA8XQEp$oBtXpv*5UlYM^@Shd>Vn^%;8z1PeBw(cQvs62hRKN za0mW-oBZdu$^RL@=D(lF|IQ~;)wiIDNQ^}r&+Uijqv{_il!vp$cEV*!=Lb_@}e6fiJOH6O+{SUvf6gJtTQI={u) z%g3KDXM{z;kLo@-*4Mq~R9GcSxZeYGre=FPC~Hrirv$n~S9#1*DFEj?A;ULZwToH3~_~idZi3 z_8YJ?mggs7OijW2IdJ*1-A}+emnTaI9EB&ZO#BUjcH^Lggv7-3%FJJz2N>6Y?-G3T z6cUP4ziRyMT@JW6C@C*oA)QIqSEImW!lY_T%nzIx3SdoIb@aZYr?vk|&zgAy;O3IS zN+;RW!{BkdBivO)$Kvrp)bS}6D7dUG&(lj8VfO*9!5M16bCYz^Dg>&w5IiB9y!U;F z-Pod|yN+J9zvKVLCH@dVvpyIYd<)s~Gy-`GUH1OQzX9L>l0E)^I^1?AdpSbKAh~`R z6_njc-VtY9s*asE0CqnB1&uUW>BiGuzI;j2KrIZkVm7U>EFS&7P^$IT0pkBlcJK`S zk@a1vUR*>TM~SzmF1n4SL=>w+oEqvOcNc`is(1)ull2>6Jwpi@c9I+)jQyO7ZSw|T z^iIwfaJ`?HD8fp9b7;e6EqND5I73_^T7M%hk##iyxN^z3 z2hr!4hkIG%_bm372)E?FULW)VQ}%rjJX1I|IRT3V-FC5T%D!JDZk|WfBWs{3N9M2Z zPAnmdO;*pg8U*)~`b|XQrRBl-$7T}u_}7b{EKv2>!y3vNa2G{H#zrbb^Gks!=;EVg-Kh7hra zVx_XaOsVb%_x{5Yz`NQ3`ca833!hqBhNB~hzEHQJf;ICqax9Zt^D&6S;ptaN;Nx>d}a44egK+JY7b+{jaXtVgyTm*hyZUxt~*MI?tND zj9+>$QQO)L@?ejL?8x2@3qX)c&~JAWe#zqft*<6(uCIjAx8={M@*pkvA_o;eurL2f zCH8%YGA7hv!(cS}|AS^s1L>5M+5Hd!Jkh2e4HSWRSQ4z{Ds`PUknJT8j@{b>(<0G+ z&LeLQ58o4<-b&Oxn`sGOtA_oP z+%{uj<18janx}-wO6}7X0e< z>N}Y`-|qVPZ&yx!b@@i}kBsHNeN&?P!#S2FbLrwQZ|i-1R`cxFn@%z3{czy3P15@p ze^>w6mtU4^y(-tsQ`l#ERm+5(+tpAw9Qs_!t}5Z7D)&OBy@7p?4lNc#JoeU8S5O8* ztlWvJq;%6sUsZXj0v+SwIV>V^>9Z!t0IDn!F;h(^qXd6)1!;cmYu-{v3a$RqNwUeQ zg;|hlT2YrejUNtTl3D%wUtR#Y_Z71M>2KKEKrtOTHT3+;bu4zNTf5l78@w=Z)OYEU zY=j?+Id7~y+`!SXv8I>mi$Trfyq=qZJ!}hWrmuM>;n%rTXr^{xrDG;Ghhgih+Zmre zT(?COX^RfA;`I-NIn-?F+7OG@C3<3vZ$dHCmBe0xEXC4%k{Vm~Q8vzZ_+ zi*POj>J!mb0AgH^r1`btb7w$#``N3PnMm~Gbv3yGp!Yg>tvs;J*bOwvvC0#4I2+;A z%s@OGf>G!C+KpSnupiv!6pDY&zE#CsQR!Ur~r>NeIy4xgHacKOUy|}TYC@8gmkk9 z8{~Lj-?nvd1|DmZn^|Pcq|Ya*n|>Ht%!hAzF}zv^&zKla78|x}dEyNrnp}7Kn%=={ zRQp@=L2syiycnn2z)Lh+fBg0WM>|2t+1#00bU^4H`|%wye!fh!J#GEa&wJw^zrVmm zi^`8|&`{~WX7P5+H=lpr1bd|s909aUd#oCsd?&!W%TqZuNB;;b2^JS7OCt40QLrIP zjyLv?MS86fs8g_XT3wb`maS9!<44{Hi^tr0N=*PbMo&gN;EF#Ban~}luUyoGx3f-F zE!FyscQsxgR@*GJh6it@Bd<^HY5iAUI$@P{|7hfW{-A`p`_+Z!y}ZAq`aD}O5U-W8 z2kKq#<9q@7Xgxi>OozT;kjIkL%R|}vD>>e)_bYGTJBU+olLCi|!t{vjNZA4Mb+B7u zXLwEg_gJm5F_IH?SzX`AyE2@e*fr+NX7~Xki0mwg|_ZOi|(<+O$=C7N)c{-P` ziu5)T_rby*|JR~zUQbziucTxQ^aX6l&H|~7>zS9#92}vs@5WxB!iKv{H^g| z6PQ948v;O#c}-$mL;B;#yMg6P$aO-EHPi;Ma#PxMtHw-_9d;vM@0T$#`7~K!!0m6W zC@(L+P{BB=)HLJr7<-pkkEf=mrmMhXyxGKfw0^sPq09J$sdDg{Z!d7M+Db#!L(425 zL+abif-O$-k9~d704DbKmpnxI9|dGj&$gkaux?_MSvBRV;ylJCB@NI!wvF=dcTVt0 zil}b=BjxY&t<$vi66zCrjLV9#<-;OYnvky}8l@GB4RHuQu=^Lc%#T#Bgn5xCE4{eB z7L|q+zwFBR2*{~(SNE{W3}R#Bn=|qD=mXO1t6yUy-3`Zc1alm@vN^hM3;Oz)WRh0< zW8d7Sq?ihQWD5&QTT*sQRy3&YLbzLkm{Qmoe~pa_sCNi-kuz#$JYrTVDQB#1DJKM~ z_e`6?%x2Qq*{x4nyoFEC;h*3WuhwV7=Ud|AF>)^yihR(npZ-3<9f%+UVl4#M6reXl zki|j!&Yy-_QsV*x<4f}L|25XKhNzifbaZr_>p9D17N91Hm>kISyi`x7O(~#*>+9zr zC3fi#&(+Ax4Bv?jX&|^lMaXRI``7~TZ|U$%jPTWx9Dn-^ZviUrQ^Yc(ZZ7I|aqCht zP-LHot2(jW|2M%k^peQ-)h-JMuI;#)^~G8d-FiMAkT`mL>GEu4Wn~@*ZVlsCksgQ@n<*Y4+Bz+bXtovOsWZKH|$=Rc8V-*t4?YVz)%@dU)F@fUD42>``70P!j1MsS9O98_4&Yy zVU}U(&}d(SBDjqTY*hJoV87fBc;~vZ<0UmpfxW;O!%x0ZD?UJF+I$foe|uNO?R)0na9ah7$7yp4-Jx>jlqMJjJ&4))`` zu-IgMSFRV8YvBSGnf&!C-~b9{0A8Ve(Fy ztr8|}2qCK&#(bTT1H6lf*H}8v7Uw;-YZ^}P^x){}$G0ie!V;mk$7%3gbq@tgVhTiE z$j3wP*7@af!U{iK;SWG4S*@?&(EFB5F^Bo>Pb8wrXz$*ya!>PKyuWyB@3JNel4#7Z zyqSfXzSyRI3Z{fGwNp1AlB^O$DK;S~H zBgWce@%GVQcL>~T+wtb5XaI;n@84cDKEGVbaYaK!GTck|(Lfa@ZBPD=<9|7~^@x$k znt4C6X)p2Q-m%9%Y5JZ>=SIlHWMw5!GqV6uAcPui{<*CQ6(V0p(9QyJrOcL{of=N4 zp$>LuL?Jd4g0_LuufcwVq#Uu5SM9>CB{`YQbgUfO58=zjE62F-N5 z&RZM-X|uT3PXHlm?U3mg3QJ>G7s z41G3#0FGMTx1i~h=}$*AuQ$2K22oEmr4J^YU}{IHGD~~#C+t=M^23Ugvj-iRp-h52 zY)(m&91lKI_uepBqtI*3Vw1aS)fSZvSBA>EX-_#FxJo8v`&KG7Hdca|b~7x6vV72WPaIgQ~pRQvqxUn@JsGe+_G7 z^<`3G49sxrs>>qwOxnJ?(q$ZCIh#pj=3ifp($1HO*0qE{KnBS+WASe8f~M`hwQ#*&V~9QAE64(c3KMKy?W=a9V{|A%+&+!%(c^~-P-YuwgsJ9$OFaYfU;3ns< zxS3@G%EZ`LS)zd2A9$@X27kFs8T4wAjhn@W3n(oyCksqt$4G^$bz?tTa*D8v(7SKo zZohE(`V;^BvG-w;T5``XFE1jaR0BRTNZ#FL3?RQZZ*~TN%ev7B^^tN-q8=G;saN>m zP2OhZrDoO;`ZPt2QRyir3BqI2V>1_fvSlvqcvGhSL4-Eo)9#+KxuF$LyI0Lku~1a} zlQP(LVUAL}|JmF$3-*`)egT%u>%UEyiMG@6YW@~J{3KX*5 z0@UKkh-|E^T$fE}iaOS5{s&5-*XT9ArIljThD^n4GKI$=y9C7!j)MZr-Q9hWDlg?cYBy_N)-($c%wmVa-IH)$k_9i$ zHVy6&F%JF2##VT)T2!$=(e(;A! z9@@!gS%-ZXPtCLiC+mZE>^$LnffE(AnHVab1L)n;3mL6!C89bUv-R>krq#y{ElGnx z*KSStz4`noW)epWZ1J%`3MP^$F^WFZPZGR)CcfaX-n`CzRc7`%ZquX#RpnY1f8IoR z6d9Uqu{^4edOF-M%#{O5X4?lvv{Sbd!56IpgTG_5oB?pDyX|(`n@3a3glG}mV6=V> zd@6KR2w0wPemzxq0)P`tIlV?-$Ub{d*n$+xEkW z1BdyZX7;IiP8_G6A#Z^Nb$Vz)Rny>ByXu06P)yVAdv`T*+8r)X6?{Y5g$ksl%)w;E zS&#u+2Iu=#qL`}!GlL_xOKC4XpZ}LWAaT19hGBV{c@0HP^j!3Iu4HH`H)h(yQK^R; z(;TR*nx5Wx4XP~|+MES37e+OBxBMAH{Q#&$Q3HjXY4O&q&4amLpoYLejIPez)erG< zZ-IGURhSy`L?!Q-MuIvGC9)s+nh}2)biXL-xW3Na>IILNfo&?v-+j>3)Ksglerfi= zvrsA9>P{cMLl7N0c^88cR;2^L+~tPnQLVf9cM@p+zhBAq(k@B!Vn?W|MqaHB2}&xsh% zLYl$i#cYg|NveuGV5}6H?cy*Wikq?>+BjW*ZrSp8Y z{TUEuaD`c*it zQR(?W@YLh#!?e)A`5{PVUR7m0*&zEK&4b|msNegiuF#Wep&3(bZHf|?^0jwAx#|LE zCzvHRv#T<;LXlH@>f&-z{S!tes0Up4$)iSNok07(I&lmla7D5(aYWsZ!d7*tdV>Ox zAyk0Wygk{mrh*4{K7_f)`Vo+`--p?Qn*2QzkB`ALPrYmZ$gkRg-)Ns zyt8MGG00dboo8{s&S;+xp(YCIZ^guy2sFcOHkhodsopx6OLvZ6`ByJ^Q%5+j9Kp;p z3xni^`?Kg-AZz<(kM_0WfOfsR%6PEZdX}Ar1qEgMU?t?3ACc(Qd(xNp$tqpx#`H#N z734K`7h8MyLgU{`1INf5KxBVO6R$QjQ&bN7U1zWH;NSbsEfi%OJWu~;vrx1$T3qTP z$EbRafzWJT{Qo7aei9NvmAAleJ1FD@1O#ATnk=&R?2bdg+qZ;~_R3?S*Kbp2UOtYv zP*(_pY4!5bt6B?Ry@aFaO{2Bjw{w5YcG{IryZw&MLFGRC%Vb;MGR{Go1*X??$ z6L_jzHfmpL&|@>n@kgD}yvrJ*m%y(RQInTxq7cUPY2TY%kE}!;siC;FTG*lj zGG@gaxvBZPD`;FE&P@DZCg7qm8kt_H$n^Pv1^$K%Ov4t`1VWvWXfc!{?;d;g>`t$- z_fhrdTNOSTE-q)JEnRetdYq_Q zM!n9HA;mr$3sWpsa z)wsv))8LBWFTK1ULFR445S{MM-P$xnjKwC)8!&v&J5z?HMjCe}PR=_hy}_nP*X`qr z1O<(_DqQ{+(&Jrkdrh|sV3HaZ>6+3wsq?2Fn=BRHmQ>zEH7lOGi;_~ z!}-HB3?U5|AXOVoIQ!QuM*PXX^RG{6a4ZB6=Kl;jMBMB}PiJ>!Ec$g+CTHhAUBAld zf?7QY7h+e0SXSU&R$C2M9#^HdSt$sQwiqmG2B>d$1^VMc(q;sxJDEnt(da+&VD#9}J^Tgos`Yles~l}IxqH= z^wO39>RGlyVh1+o;L;>Pd>gK}0OS!{e{M?hw8l`sk6ztxJM)jkQAff~PQ#u8og#_U z7v=8FuthYFw#*v{fBJ=A6FgEMnPhed(8=#6Qkv9qDAEf5m}x}PJp71s0XSukUXKw0 zh-nYDJnxc2!}jIY>bR+~jT=|7C-z62_Q& zfpDFHaV0b_3e$?+pf}lKZbk>13t37^N_0QemQ0U<)_$+Jcuq_6^2q9(iuK9D+M*sO zxhc{4A_FfUdIvSp51;`$1L`_;@GEm3H;KgSA&n^}0cDXrLOqUT#LFA3?l8xpMTHWH zMs4B4jRPK>QJorJPLq2cd8p7lcououWLmR=q!=22c<%SlNpR5N%u-6F6HFatH0;^G zgqTB`S>@an6-WmfcDwtL&Fdr6v%#uS1SB9qn;8vxszd6JFgz@;uHEMshd+it@r3Dw zMNMQURDXgC;b!ULWUj+_8g{C{NMgy1SuB#bRQ^WWBi{Qy{EE8+=IOX0R`Lp7vmp!? z`j#A}WDdeC+^CEUso8z1Sf=G3Zy|Mi)?p<)B)#caG@^^?y}FNeMRF3cr$nf zA? z_w8cs0V=63qUl@Idro9?`{!!xUYrkd3!rnbKGWzlo34m8beGTFB&d%I2P%zyCBmRf zrm4zIoR0DA%)l%S?XZx0x9DBibcg4RF`M(DvmA~ba(d7}J1kmaQE3avYO9-B1AXuWV;c$=kkw1+7^%8_ zSQR(=cK^rE^dp5}KiSpnB-wtiw)~iO^Ylgv8#`S z%Sut=fWzL~BC{)cowS{aL!D_%_1crdLy#_0M|+cLJOsbTCa=+a40fSSgOg}|NpL#; z#C95Zy2rHqzGHTh{kXFmdS*THE$cG#w9O9YRg`<>z^{`qlF3nMg24tl9B*uDfP$z=51FRd(%4+$|7%ZQ-&yUnCenQgZPU0a zvR$xdJdDcKNNlCRg}Df)gNl%yN#sDtGzltNw0ylEnT=_T0|j@oCKQpKXbO`WOE86( z*snGO$ypmQO8+#{iIS)+hTqbx!@%)7!VF=~NQTqNo_RDeSect9qTrL%tNg09aUiMA zCPG;H__$!sv`2GSNJ-#jz%L8sl0;>7PlK{GaneSL>e;0aPHZQw2)3#`W&zDGof?Tz zDCXpjCRBwl*m{{#M7j8V6s7@J7Z5O`q1>S9HrkbakGjxuD4d(Uh1_f8&?CSG-_(>% zb1%=Hf`?g5E;$xBa68j=^Y=d8VxpUwTQ~;xV*GFaNvN)%y7iym*Ogy{8Sl?FTSIH zBf#tIrBaSHC;1jRz^go7i`TAAeNpHgRA+|Ve1vD)rqZV(c1;BYm3s%aiKhQ_<R8QWFc~2WSQ4>9-^DtisXE&CM-qD(vPx6!$+!eOq zN9KYCQcWOp;M1E2LzV%QJs+Ys>Jb?%t%jdh+`) z$!VePG!6GI&8AQFSSheSfyU5jJdrIf7te3Tz^R?K&R|7+nWBh5HXI1r6K0}sj*jG> zGR!|R@RATJvli|gH^r6J-_I94DvkI|(3Qr80GTscD*z4<`5)Pb012df^xAb@za2$Z zO=@aSP{4MG2XjLPG%*}_vT+qPw37uC|AjKia-tz+0OEjV>N9D~NguzJmjsgACQNar zaWQRZ`7+CKqtVE)5Hv>r>lR6!;`t+nF=*+hpdQl0;IZ04)C)Ou{w^RGxruztlzM1r zGff|va}$%cW`j+~EU+ncRjN4$slJ}Z3RM?%;2#43jpCVRyzR6>S-_d-bWP}gJSW5F z7671?S($Q=95gSk)r=uK1l$1T(-~1wh8YX`FyLnwji}rch-~PF?O(cDL-#h7l4So* z>52|90(PdQNI^kKlfGy;1dVAQ!K^Em4Go0ANQrS&TY)(E9hl>F4nvZ)OSj4(tZ0ny z>{a4F_jKNP((FVM28enV-1CtS)M%jkKm+2i zmgOkusztA8`KNu9ogt_|{z~dFX-BWfyYAU?t?B8Lt{7&iK?96<0-|K@$D7O6@7;h-;WqQw zm+OA^kA3~iLCKXTjvo74(~17ScOS}L>$fztQ}4*Z-xjw27H1Nvbn}&hUDVmPnTwyP zuiLXetVBxh)0wI21})3{=Bb&I9{aR`VR!jb?)#@D&z*+{-N1{4qnXCIVVVlhzP`Se zIt)(mD=8^iwrp7|=GqqEKB!+9RNCW#o>KWq3vJt=0_|Om%|~YY0HM}N9!gLTd9=wH z&rd7Mq3@TvLy6a)jx2V%5TYNYw$4!wejMOUeyudKlQdct6ViG!?O}KB>g>ey&BHd> zj#Hmne5gk$#jq%V4CYcGX;i{xQ5%)qxcS(kHhE-hEIKbw0TUn%<*`5wE3g`d>=to9UUEL+Cu;@ zv-gf}j5GmN>5HVc1s;QO`0w?)`kOlSXU|Q?uGIW=bo=#xOH0iwHp8w?(yhWb@Xo z#&i6~fA!ybk$imn_HDF<<9}4Wq+=T7Q{_XXNwW|_0^#C1lOA(kugyjZ|4Y-hOplrW z=L_~%Ok$lYnwrA-0_nho zNjVLkB6?XdihL)#X=eUlXizm4@T|6(XLB%~ys{CLzuH`P0PK!%UGxJq}+hAapFQ zY7efHX=-g9kvxmqn$3jw47=hbUbAa8ZPg>cAv^TGZiw)laW0&B{`@(M4Q?R{U;Ni= zDj%b%4jwhoO}#Sv_U%iIUir_@n7OB&pkhKDMr>xjH-8c8lf9;WgM;T5E?mfFA|x%3 z!*g??f5>QW4$qN?oXT0kB{=#xk37G)Nx@YU*bQXbJX3>AZ}Yr+9wSnKE>C#?08`bb zFSKE1W@ZW@XB@_NUnT}f)n++bv(2;4X~b*Z2A8o5v~nR5Xkcz`ZoKB>(-z!5_;zR* zsP|2|3Ks_BK}=NpC|yOyK;Eo8nLVGR!vHRWv%ev$iDPX*-DjCCBO@cN@PR++&*`f6 zxpCtx;0`M>>`fNhBr?yS{ZY3U&nL2Lr$MePbZJFnV`FI0>{iU%X_=tG`WbHHHVy-5 z3j@wiNJt>7Pcl4wc-ym1PEO_LIJ9)<;I{L;c3oq_kW}dwp+B;T`8Em T;pQL(NY+7>9}@q5>gWFf@{4TQ literal 27085 zcmeIbcU+X`)-F7@7&Xc^u?v#KQ4|Cs2ue2^HG={MDbh6xNRRXm(XA#nbP!OGVgaNp zRbVK_SdgYvVQ2~iQU|HhziW+2-kk4`^S=9h=bZ1H-*5ZN-Xbu=^W4vUuXU~Ky4K>? zgPQvmFI=;b!C)+AegB;{gE7yY!I*RO(@*djli8J~`0JA&_wV}-f8)>C?p#0owcw}k z4V)MZk#qFFSpmHF<_yM{4Aytw>bT$RY4p0T>yoF|KWeoO@B~5?t{C0u+Ec#%fdZFN34<4LN_sH-+^?Cnn`e4~L|CRK2 zRMfxl7EKGqgVeM6`4f(>_N?SJ&rCI^b;g9nSe8fB9S_t<#TVLtTO?hwTS%p<+T2xQ z(TXk4KUr&g<=fMrn0Ho0OJ?T2f4cUNzJAEk^?TFno2%ju(LYlBrn-Zqg507iHI*b? zTwLN35{mOio4D0Horb%tz!!Xea&BK&mm`}aYlBPo|N4u5y`9I79ZSv|ssBtwahp=l z{oQ4X-H8pZTsx2VPgbV=_~blia-cYSxW-7!_QUF!HdF07Dz zvD)(HtM-v-?ah(nRj~u(-6r{{3Bep+{O zuEpes_a0-skU-{~`AbUPKNF3NI}~d%KGZq(NXYlf=_QKyIOVw4u&u|Qww;$qeixtG z{DXS5w(p0>q89k6V?1fUy7sAL&cKV32oNVva%L=&nef}=f=QmV2QNt95trI=H&F8h{gj)IFE)?$I{1Y>xv;Y1_b;@z@ckN2 zKI7FEOuOg@&0K!p9_G+qbx5o9MUVuSQ~ukTMT1YIeYdh4t9A9Gb8^PwldCy=2v6(S>N*jX_rHH}L)FKV z-NxQ8Ww5?MgFVXWP|4;68#&HQjat^4xRv7S0+~X}{X4t5yPe+tDrWKM)Vv$ptSZ*L zv1j*i2FvQK6O!0Wv2bT?vHZ38*^GBPW;x8?Vv@5hP)tiIMJrM@uN2>^qpQmkQyXJ> zOyzVZ7f+52N*Eq<8!2`0m1R}LYDM44`wM2r>r%~ahE9$)y2aw1rY1(E3*;S|IF1GMrrn=?J=jrRo>6NW##EU8&|0fF zo*DP_>C>KqnQ1z{PW4u?(fb0{DZKqI{=I1k(rrdcq@I~Y!Beh)YiDXXFg9lqvI0NdMs$*@o zsYM&>PTaXPFjnrTDr-@4xHWBm&5W|TxR;ciNQdU$>9fJ2f5%-4-`lZ{pUNvDZvPZKk@}N=y0ucN?`LZuVw1 z)1xZ=W$AjElW%{Y#6vlJYz9YF+H86tME1wue;Mj)RJyY-pscMZP!=n?qT4}BLuAU{ zs%YHQDe;0OyyE?8A07!ACfOm}M0@e`vJ6DiEJ_VKysT|)Z8NnhTT5=pT82k9hU5=L zYqKk`3hIdD(Z$jQoNITh67*9loXSgPR{5xqY|zPt7MvCEqs5PUk+*E(?CubJw>KC%(c{k-FcIGsADxK59B=gx|X zc`J=yTH8f__+6~#gL2KY5X4-^W^C4G_2v&xw;QTVe|1hcwWgJyUa85G_ZurVtMO=9 zEo*U&OIxfrK3k=>6iMp`-`+AMzVV%uEYoK=%k~+~4nEyRfCf78DJjw-z8~HsW#(ah zYYqLpY`UJi71!&$t(OQ=?^fQYKcm!oQC1xxv^$Y7L5;rMi4g6$Yhe!g*_?c?eXdK_ z`CTOund~UJsUWL@2w9aAsyJMA z$O)%J%s{SdcxI|#Cd7AYoQKOSk;chOG|rZxjEit7gY9Aa@Xc9dfy{dGT4N^-Q)C_Q z&i!WUsn+#pBBzI9dHKzX1$~ql1_nmB4$amk0jlFCVl=};nPb%{zCrc%^#^Y5*t*ip z`(%Xs+h2vfdLB?@H0*8tW&P#n%`?-Z+vMfj5d@4z>MSGWeZ_5PHw}jKy<<7~gJnt` z^lgPd&v8N^=s?D$xD_~BFf|0n3s*ZJDJy%1_d=+y0a%?UucHQGlk9Sidtyy)_ z_@9ICF3Of|QtP*+82sj(XoP8=Cl6_`Z1MJnE6i*_p;Y_`%d(paopouJdjiyaCqE46 z2v78+X)RHF7gbeVZT|G)>TJAaAhTes)M4P|ZXsTi7hl7~#KfbS6}XWXb$W?9+t{-+ zx(=JrNTueYM$zIQ<@Z< zTa6sWqf((8_6KiU8$JB)PJnCu1B>#A#-786Vl^d`jI$jtxjMC%Sjo!_?l&8+SYxN? zG}nP$p=V^L>p9*TTYF-3pjbNO>=F?}dp8`QOi#|O-RCk7*8)ux_ddO}wzkXbaPOn} zqFXd>?s!mVL#MuTs8Y+p%&QW7y?vdAEj>`?4cLLqjWoFmm>W)tY)G zrPKdy{rrB$QjVlgjE3%jvUrf911>;2V4aG3RTv(oV#8e%i#0o}QWr znSLMOm-wo$NGx)T9l|Zo)iTn@HLbTeSW-e4kTpx%l=JY(F?p2$H!RYU+NR*{7?nHb z`ex6NMjUEl9z1y9JWwQ-a?+IE-o|)0z3|c+x%dVPEb{WWI4pUzHZM6ZMy(Y;F6F&K zl025l!YAET3MX^thU=%Z!dFO}HrVe`cC*lfL$w zBRlQ$hHE%8Q+x-fW(v*$n`}E@?@dNl(pgH<2<(W{bsy8rhutSPt98-?mLO| zU>O_kYS=m~nfdzbnL)LgL#74&zbzVkA(3*e)@!&cNZR|&BHnPAeNNnyW0NCU1(MQ{ zE;R<-Mvi+LaYr`Zv2s2wE!M}LT`t8gu6=kGz%K|$B&gN_;N*N&)hm9mx~UgM3ydv(qKK+ZHQSI)0MQdbA4+M(*TEzDMWfld?uu}d5yJe%xNLj9goj0wmCWR`oXc{VoHwZ0xE~TGBR?=b2&F* z6%Zqo>WSK;SZKCQz`k`JOf8jR9rp36dq31U6ilI}K!P+r`DD3W-vrK~cS6UJt7Uhl zM!3?s{Cuwq+;c*r5r?!?GC0$@%~Q#@pQ!<7qykE0JoHH`m6y=^^~zmkyWgz|(0OEc zIx`$HJjX9$uY30#6YV~cz1;2jSLf2*MTCHNe8z`1r$hpFOKQ$}H5Bl5x96wLIf;C? z{Tov9mP_be3z-=RDM%dt>TJT)yM1aesQ5@o&JpK(lzSQ4&d25p`m)6?HmhvHz|5?9 z^RuyTvl&BERRJb?JCGDw#v4m8n~FcwWf&HKjge%}uT3#a8?L_7Qg4Gr z87p6|GL46wRVLs1EI?X%2B*+A(+;c1bD}s|x=5OH|LJsq+O&Iw_dpS|Xo1+xgYP~5 zS^F$!?-K_ADIPE8QJ2whFE@+Rux(lIFT+#zacvn{y_wwL5X+M5SA9n6#N3)ZEQ6&? za>{%OcgQYq=f-vZQg-*)&t^uG*yRo$W4lGi4~C6cd9EECELUr`dgEmC%p@%9?D?ci z-ZrIOSMRX~@5iOt;lc2DdoHZV@S4Fx%;hLslw5boOi4;AefR1Pw`w$J-BZ20gZCu) z*_{*Ns4LR!jy=7|LZnZa303{2C#haWd=kZgnTN;l<8-}*`1lt9P{rK^fU~u)XW|o* z#ib*E`fWkCauY7_IXA;}bPNb$IX==G zCgRqxwRxgxbN~lF@ni@+RA>2F4obs4b+RagBvVq>UDCHIR~|gzk)3qZBE>W>G)dYh zGx4@u#_+>%^EbbpahTW|*nYHzy+|8+|Z922k^x;$Q-;3gxa$_TXJ&`j5 z@l!jHUdq%EnMYQBhFcQ(vZdEw%ctuQkEhi4@Xq5YW3S$f8{Zt=1-2kZ&c37X`KH>e zzRC*W8@;`jCW@XnN31<>W=R}T4-GKxdG2NB8RoffpLEu7lVN9`M^#g67SGbKvpl83 z#{me1XEm?AG%V!YI?sC?yP8A;9(PPkBXlU!WV>Cp!Qihc0mD-l#eI5pQ9z;GD&8-h zcTIHW!?U`h=R|5w95OSDB#1QDGGFaFl`gk#9nA|M$d$CT-?b3H!PGP#m_kul*x0dX z1J!yDYqfdb`?Vi~w zIzCjnBdl+um{f?mjhkLxsioG~*P5td8L^LLULKuUVqWh&cev%z>98}u38@Sn1gnrR zV;E+cCAmJx3@bI@SQ)1jsyH=NSzN3Ud;B2|=8NKU!j8`s1D=d`C)zl)UfN!^vf#rL zz-6(Nyb(hJ6bP$cmo{&pB5;w^!zdFuPzpP}7Pu=N$!c_!XjbGLd3375qa(bxuQe_< z!rFBapWeZOi{$(aKhQ(RsB+u&RVdA^ZET9WHBR5(-t3)L?OWs8J*M<+<(QTk;9yqo zLqgauitX}BbK``3`%m+D^F_UX3h!OJcC8Z?UlTiQ4s{N0>f``h3l~d{b*RlCZ9U(o zTG5`*eT&lD$TAEyXCg=0zTsg;=2-tZwVSFw<4l%R!Q=qJQDo!d;aaoEeU;B8~ajAZOYly{mmI+_m-b|y#zL6FDuhILKTe~cKu2d=N~+#p05um zLA}A{_{HhG`2E)}io2D1(<|59nH6?c9~g2>ai99eBDq@QL#C=9T26=A*{s==H{O+y zG$QNP*Wj`)L4wk#wDk6HqfFT#^=OMhcUR_hF7m~TVh|!cMQ(YxQdV6Oulr6##}5@A z9Ti=SbUA%99~a~R@2sHyXKK?Y!|g(+Ck@w}F)}g=OY!DOu=M7&2mn`5ahIKQJ~ARO zklkg4_UpEIL^cQFDysl?WRN8W#@{}yR`FsMu3$O{^w~xOIkTRwYT9e>q7WW+bxln>qKP`tBAc1j;JBZt8q^AW$)?Lne$$iQ?XrU>lb3}fHuZ9&4gLgjNS`j4_sWQl4B6&GuoKNQFiI7ngqFK zbSDx;B;IH2K-&D>c_Mg&7sV)-tl!`I4kekL#~2E*-k}_J8^}XxX}^KM$ex3u)z#Id zr>gNr+x=!HnrlyFf$|h~srrGkb^Wc@bmty*gn^Fm{Sx}gB~rOPY}&$?>Zo|qmZmkx67ctUXe{or>GKf1ORPTNxgc>#y7su+;NNJ``ZFBPLj}>xeP{2W2|eV z#)%@JsezW$LKY1VPkKxZ*A60-w>MAsHp|$)zJF0vO%?D+6U@LMXuoz8RQB@L+gYeZ zE4p?^j?~+v+(hEe23ibET(}Czfg_B7LUd6Yh|I}8w*s3Lu6?MROYM)D5n5Qi-QHNC z;RBufsnKS&`Rm+|c#ibkK;7StvfErQ@$gl`UMSZ0?m7Lm1~i!Y17-aqItGvY9Ypu`4n-*)Fvk;PKETQ01WE<^52JMrqa`nAZrse-eu4eEIS z$kVb?V|(e;9gVL4ECJLM7gkYAF$xbhx6?6?3;SUU3wRbsw#JcyP4*r2;FEf zJU{Z)oqdAJ7|%%?*=vGdpoYLL{X%W&Mht z^+OIluOA3lmeQ#iJP3*5&PvPhsR+-mxM-lwbi)qpz6xtmy41K^w{E3_vI=2e`(*K& z^qTzqfuNLYqc`MiY80DyR~`d}m-t{R)F#=$MnzCY*>_dN@$CMfLWVO+<+<0`*Q~9r z)r&)<%~JB$tX;dmBht?|A#w&t^N{D8Gb=;$MzDM2yJI`EUBQ)RjeQ7G3{a1L5x!re z{0>5ZQ1oLVHBPmvm{x?bVYZ`rSo2_xH>ehS*JKf3(4=)Uz~A;^0W-+5+qiBB^~)2y zGF(iyTi!gmcPs-t&*{<>*GVKmSRu8+_0{b?l9^uE&c(fS`k#)d*XW2RP2@)F#5ZKr zzJ7gpye2EZI)*<6B%f4Wz5wed{>10b@^6k5`WqwD8Kj^1A-Q0BoCpni>+w;3B+mb5 zYqo6?1vqz_9_z3KrpC6_L;=HXpH`dd)C$>Wa2W@Nc;rJSCgDHNnQsy4Gg3sXGeE;; zVnB)y?M>I|0GStrEi1WC#6^d~I?9EM!U{4}gpqnc7qC3{3`b>&+GKyp^{wk^qg598 z%?xyFi2fAJ1oSz9JahV@;$uQLRMW5=K{dQ@HJ3I|70fzKICh0+lNqKqQ# z^aMY0t^LEVD42ks4YZiZ7?HILP!C5uy15?}V;M4HAP&lUK>Tu0cNIAMJvhAC?`~fp zf@t!Y-{ij7#=a&{!Nxwr)i+eUT#2*?nGp*Lmd>`+!5`0j&5AlZ~uJIzht=shBxj;pyXwDF%4|9;*{A;b8YM_8h zFo(>9UFG1)IFiqL7n)6V={pr|2w}1y8wHuT)6x#wwsXe8vV%zEG9-8b27NJ}N2B4NpcAe!!RCMoX|FR|eirzU(D!Gwplh2R(b;VJ39_>T%kw4sp6)^<~=fh51Q|(b>V3{O$0`lPv zMHg#doVkrG)G89k#eV^}ffTsWz$I*k2>7%zMG2M+kee-t1`~s(+~7a%#HnBVFqJiq zAdW1dCzUp2I~@W7zv|0>*m{{r*UshwKZ0#T-L(#R6Q~fy(yH~0qQO|PWSrlhKW-rj~f%b2gKkNhW@)gVKK6$vH-?+Vl(BHpEeXOau+aI&LQ zNJp4b>XXY;n$;$=o0zLXsSwlClOZS>K;ks-S^gkYaL;`u%N8%wLnUpGL=5(jckj;S z7pdmOHM!-)m3n~Vrz5Gi4H^+Ch=9Ju7gx(VmTY}S%nnOKLj&<76(Fk}`dtZ-{2<7) zMb{U;Lk(|zcJbO^A~2Dc%b=!7Hu#+TNpO$hs~`QL8Snl7pcVhia&xt-%d9=As34N1 zX5G62z-+TlQ)PS;y?#yk7Tg3tQ(Voo3~ z|EtRR7XB*7KfM6->VHW^|Cd+hKh^Se?DSV2v_iCi((!_}21!Dl+$;xBs3F8%t+h-4 zL0`ZGFb-~Qw4Td9zg&VOqIRX`F->sZ4**RSLFYdQeRK0zfQo#8^1%1{A)nH{eS85W zm~*1OJ19-aLI|p=s$#O%dG|kpbhH_2r#*^7Q(r137*Wr2N@$ z@Qadl)jjXF^`bp7pQd$<7IXQn0csJD1cth*C5UnJMf_|BBg!I`2_;|~_O@n$?^I6C z0eX%9`8Wii=wzuMeLzji5H2K+ieP+Ebcku_rR_~SjO??`X+gFaKm10;YovqRis>+9*mNW{s|3^22ML-k@-1S7W{nCCZpg7du-{bzZ#e)V&g(XnH zzN;h7L)leM7YHK-56qDD2qQB_@YF=r@}IO$PLCEi$bAWsxh4Eo#!S|nmhfJ{sLrm> zn_Tw?Z|UssnbT;or0KZsb)}=f{nPPb7!y%*4D4U1%E;^W#EJG5%uGuq{)5&i$}#7O zFb7*Yqv!inFA=ftN`vfu}IT>Wy8ek0=2e;Jbr<@a?9Z(OXS{d6ck4iShW;sZN`cCn3x#hetCKM@u{VZ zsPM_@D%D1d^JnRSvDjmS9Vspo5-A)uv;C#O8k2L${)>KS3=u50pkCn;A;?55uHx!_ zL3h5>vP*sze~D7h71{?-2fV=OU4^pNGy;Y}0ne_?3;Blw=+*t3I5HsMT&z*aqmRp3 z=9V8w1^5klJHzDc_E5y(jdX~JMj|=v)FP?xvF!RG;HhK z=YM`Q`X6x?{rf8Zry8~Xy`2B`<<#X?&s)ZtnVvdiBKgI+z9@thP6ON{*VLn>5=Vxx>$k9hbOoFVkhq(UF)z3IvFS&U^Dnj`CVp9W8B$FE`$@4DE6pCX>jz0_i7Yg+>1I|LRrX5>Ue2Ux`F>J5T??R?&T`>D~h zYwZQr6Oakf2skDa`QSNkC^%VM;JP)s)DD40^VvVQX;!GO+7M*BjEu&-0ln`-b7q~B{bymm;_>ZF&Qom~@n_shYba!zjx4?YUd-QNLmvD-P2e8+s=79(>FLwG1!}Vy>46BIu|{(_5Ic<6x_8`D0L3i0L`^dl zU>G6FY|fG$UxE~?;+hcA&)E6I<$}Lk+w-e+;L~e!dg1vffxZz*J_Quf2J+3`gIQTS z=v)0YD{!F@E0nrL64vPY&DTB$7yqdL<;2b1@0@QO>csVl!X(K#l9q8zo{tx zRc_!$4RtmZ=A_?}2b@_X3XZ=45mcCZ}b>$V`eZ0j@Oz@k7gF%~NIrqw)pJ;Ql__~-piIOQxhG0Jt0#gfG-*zl!1 z9`5m4h|rD}e`b;2%&4E+UIoD`8nt8Ra!?CtvU^ZVm71EVjP{XfZk>wfH8W8~=M(HU zNWWbd|G2qm+#J~Y2xeSTl4L<9R+Kap78|6*LxTz8{aD~bEB}5uWD9_)7(SUT1@bV{ z4Gj2B^wuU$fujm%*1_bjbVryj#os>e(Cj910I=BbcJH`!=@R_-0+~qQwg>T0Y!@PB z66ye(GK)-}!GD0yUM&0k!*D58{^rv5ST z1}jRoXt2>Y(MHtW1#awt8`=@cg0K0HbY9!Mre6J_fV6vg(qzp?*6!uWKU+QwQxx1G zk3!ioi!)?1;TF9;C3vZ`{+|qAmFJ}v5~nL)zkXO_x4q$~-K0z~i)d|dUmM7GMV@9d z{XclNt~dbR-ATCSHQ1!hDd3sb#(sP%Zb9+dVkN)!gDD}xokR|StJHTmAuc}N(1r^` z+RO_WC4xcm2`EdvuegcklP!nuM557(* znDzvfz}F<}pMD5q=X?_^qlw9`kX16rzdtw9jn(jlfN6=Bk5$F02BrA)+DM+wWjVgP zb*>#68!?B2k42~N|9v~;3AAF;dBW<+cf-=V)Gpu+qU>n1fCy$EfA{j1opr(ZicGkF zIrQRbHLuZi_PQn=LX&w4KmFq$^XASyBrEsA$hwKoX3AKp-?@FvTE=64)h?@TVP+O} z^Ek}t7*>+GK>GeBbTQq%g=HQZ***Beh{XX0b{C(EavbPWAQ)&nM zm3uPsZIa5HgcA9N{HexfT+mKkab`;+l58N8+~UzA0h}ZePQ0e&d zE6gHTj;|oU24$MSCtU_Pgjqy>jITeQf_v&`=5oj#$8Lk)cK2nq=?~zTGhW+|^wcTz z+myf48UVGwe*Bt0@^kB)=?TEO-13_L4i{?$Tz7vH=trH*!mEl?~ zD;Mr$@Gbgl~n`n{4MgkZZ{D z_5GvN4D#xnAg~{fT_%4@(D>y4<8swg+ga$KAioYM{ZTvagJaJ!4lbqOvU6;?{3$m8 zqSa`BAXC*yFE*S? zG&|XyMU-3*fT4YRMey40o4aN=kxs@+g=5!zpsk1k9<7jFIZ zu~F+wJTAmIR#SMy$8-PVvd{ZR_|0l883@9E`{7y&ylc!NGy$A{ts;28&6`?8cKmeY zlO@8}q20>kHYDE)$Ua6lXJk*0GYrewIQz0EUPgT;yyKgR-Yhf9a)f{Yw&HZeB^q1{ z{b%jD21Os-mCM2f?h`|uf#muvh=Ah?^4xuOuDnVVwdyeZgCvfhr;Zl}XF3e#$;s!e|m`(mT*gZ!``5WqxlCM8;2mLK;J+PGsfHI5p3&yJx(O{$i zbNdBx^8PTQx^VgV=qjTYkw7L$;(dJpa)i~hU1crHN?z>~4JVLwzm0~tM^{Is1V55P=z*HRpkh_=Qb3g{;>5<_?G?QcS&XR(o)XbmLn z$l^n33Xb6H(Izjar>iZO{h*!j4(FggLt--9xzO-s{#3J;A5J@K_wLgf&VW$;5^tFD2V8*7ERjC{{-7<7k4~bFC7=)_N}I;D)9N zzig9S*W!kYa|%P<9U?)7LJFwVHUW`Fn9EQe#qTCMFykBWY~9iSsqYSxJjc#+1l z%OZhHh*t&(K5IdLXWWa6V~zGT1~NfPpIp3v(LW0cpOKx(uDE3~!Y)`eb@n72pT^N6 z5GGk{Z)l)We~W2eRX^<8z@l~^itG}0xsZDR3`1KzZg8CU$m8q}xuI1U7O&I2n zWTb3hp@KaWr*mF)3vHR5)e_Eg9rlO*)`$W(sdu*ktmp*onnMTBoA1?J><0fnuf_-3 zYneruU9e4Z_C`F3&3~PZ!-Y6U<>U;axJ=Dv(pl30bI?6FY--M=PD?1ICjSwd{oUXe zrNT)T6*032)O3My7{wTp=GktKTgY(>P`l?Nb~+e@(BS|vvm{(;J926n482!STuaqj zl&g%E=<1sUK}wQB&>`66gf@%ie|xU9TLPKX1c|ylyCPE{ML| zo`_jmK%F2<|M+q=zP$eOj;~n*fIcc{Z#`_qMHD21j0PM2b+b)>?Wpeyz0j%_i-Q1z zi{1#nLm7=I@Qc7QCkq3}fk;&qIXO86bcUVZ^fA1MXGs4Ti=1$j)MH_%lC7jAWFoB+ z13=M+EbnBbN4D^QJu@_mT29D6@%N0*l0ojbz5nPFG6Bipv6O_$NsKo8%{cZV285C& zguER;MAN2W=tTcUA~gI>urdY0wPrzWkmv#-*Mfaxj`=eYw-Uc+j?yoNmb;96u5F(i{kWi1Yn7vb1tg^lotI@M=Mdp{MZk^b>Ny#5qtByT%9 zYHgReWs#Gk{v#Bf#KC_KbZYds4LXWmhcYVAnF=sxYNL4rC?Tsj!;8f83YORgRv#R^ z%4nbaZ}wctYcGu7%xuX++fN{KrK$T- z@t_y#Ts`rin~-u?6>m403BJit@B6|D1RN%ddiQj#o2%?m_5A$@+E`n&nK&p3Rlexu zWU*mT|9ePlgeJ>oA`O^CkH9yh#iriOzrQ{J>4sUf9fnPGc{)ei0TzX!T41t15n5;b z_oIPkH5Q7Z0kmfDs22);)5afj(R(UR_X>rr0v$NyulxJE=Q?+?&0qm#vB??MnB!6m zr))K6`k!BDKHnH-%MN#~-wZVRJ)8PH>rfY6_;X7y|Ga-=5tTgHTGS`bVw1JmqmguG z_{LRO+Q@3uNvHtZ<^^%V9*V+2aO^-V(lUkrhw4Q^>lgXAJW!QYBpGG(z;8lD2SGne zIIyJuzAZUQyGBuC7onj}d+3(n0n}_hqwz}c(as0;TUx>?6~O3c4&-C^N5vnvd`*8D zOhv>T*#ir-n*-V@y!q!Bq}uktVbTec9Kj-l*X0HR6>n$tX1v$!I2rFhy#VMEM^D^U zba_fqd)wduLZ=Ki@Su#q$JP!5hdNr(RWG zOqS|!oEEP$Hj?@|*-SL7nL#H!{P%|m%ut5@nu-jYT8bys5yfuvo9^(Vea=Mp-~kFB zq}Dc1wr>vO4$>4*n0 zix4s*G`2TxBNQJfXruo7qEXKUxq{hn&E4_nYoruM?O{&2^swP{X0eUI;nGBezm_h; zZ)RfxO%y=#oo3^t$pSAPK6W91aNjKYzsFu!xhb`XUHlx{QRG1VfhN2pnYoN zuc((^!dMLnuf21PU@w0-vhc?3qq+;+1oeQPS=c8W!RHfI|Cj>7`olS*{wEKn{Rbt$ z{$mQnwLi)9yMIi9*zl*kvH9O`@h_}}b#u+U^h{XoCFu-OyDBP~yXdp0_Am0_Mmu$s z*_o`P$FmbW$Wjh*dYb4LQ0_0BgVL`Qi2NpABOs)pCS0f1#iad$TIYnEuI#B%M9dOm zUEr`lqq!qXb>M0MlOUihymG`663_o3Dn3*UDWLideSUAD^5v0BW ztqv#Fl}^AupQZb&V8!M2iR5w~-#6;{`&FcPG5G8Js6C*kN*U@&0b_-##I8F7G2JIx?E5R@uowaVc~9}G(I-T-%0Qn8>M-*g~1Qz|Nh-1WtvuTv33KNZ9o zf0tdl7hCM0Sd)>LNfL1_VSa&W5cTOFM+!}MiD-v~eW#@mivgGXjV{XmABNpS|(VMaJe{UwM&vKdN{V*J{ z&dy{u!{!Kx9Qpu$44IGQVbuTW`7*Z>`XpkB&MRGNOxIqzs#z(kAG|v z2%bqbL=*p-C*83`7>s1jZW4+@duXxo{W*o_Oh82#h1}mob1sZR7=>#df+w5srk$D=x&#IRigV?dO(jz%H(_3P^(;Am73hDLk`qj3SV2#ujm zps6guqSv9hBs4_8C0;YBE_pP82D%MowRS}(uKBJ8;2;6j!G?WJ73zeY?$>lJI}_qV z{YQa3eb{Z#!5bXA6vAW_xE4bU{6@WZ7kdPAqs%1&md;9aL zm2)FY(fQr|-{jZY-v5@mx|b)X#Kjfe-gCh)FH1xESKqT;)qu3=Hluy+c|euYvhNnhKRK5d zuz8n{&lGU&@>vei+C;3Qq`V2z>h2Q{M>wXhFhN*6TQ~}M7io;97?h5XiJ1 zc|Nrd46X(HF6w%lB${}Kk$k!ug}LCs#5_j7wUQ)A`1!~3@t{il#uj)UOa$Y~3LU$f z;EGZPi9pX16j)m3e!*S>8iWfPpl@NoH2A0anVi^U*sj43fqCOkiltjsspO)lYJy=U&T-d zDK*(Z7}8ybIZM>{K+OWwKSy)I_VNX|RyR6+Z~uzYPfhQwgSLKf8H_!f*h@JF^Yq;2 zI^X~ags}U`OFwG#|JT&tx+pZ)a&vx3y$&`TfFPZ%9{qW&8XkVj^fDlP!a>m1Ck}GLO8v+?=so0lwhRJ<= zvOthYg{pAX!5ii21|p1Vd-9hc&^m!}+omSTkt=(Yt1pj`F%Om15Ogxn}eLHFZFEu6dh- z4Q~N*?*$oDb}vz2;!2yuYe(=fEWk98a_ZFb^ZbogV`&V-a|s$PyA$&B^Pli&=o2_Q z8O$tc0NG9zI0=r>;S+fa15ql|AUVvnND^Ip#H0dcopHg-H61NA$^qnmD*N!hH4Ph} zqcU>Z84CNET*ii~s_kgbT9KMk)8kb$6cHB}983q7{BvZFLZ4$h4K&DcgzxlA@pGdjd*DSNZIqjH zk<^C6>^AVk`@w~CAFg4jCK~3>SYPsr!3rp1#-qVtH|~QBiAa7lDuCK9ygQVd4Pr`P zP{k}7uev5aT*3Wkp^V_YyiA)~DX2N6xR4Mq!D1}Zere9AO|CNIL?*hqnM?a--`C4{ zs&WXD>C@dHS`gwp;X$CrKlleM5%6xnNYY&}>&PXvrfGAr4V};|&t##i=JErW$GA&Z5!b~GVIol%=!LdxscOa0|0vhY+ zM3yPc6)WW*12ro63>sB=T?r}nrlSFn796qp+Q9;slOFTt@wpBFlIM`G^~ap%+_Rv5 zV9YC_*#MajeQ`AC8Yy1{^;zzj(jR!lH_>|%-yoZG09}BYh~dy+xc9|F$MfH}KJ#m7 zoh5qHf90k;(rj7uD0%G5Fh{B6&w@Z;Vg+?JzKOun4YE^G{@+};gAkP7uJv3|e z1{34FOj02%(-afZ<_2)}`3@@7vk!C-Mv5(Fz&(fK?Ug!zSyYY+4#rrQenBYBL>qhk zOqE6pwb2kI~uk-GkWZ;erBXxp4-@$damU|zH8R3 z8F+g^Iz~H=bclrXOJlA_du8TJAM}Lc19vnFKa2WFVQ$>WnntH;2N-j~lbFP$n-jiW zo&$YZd^%al`Qu$Zn2J=YX0rw&1d(JE{764`nxHO1h(K1Ps~JNvkVuTiG+h!X2C>){ zcobx)`If&XDu<@;(Se}MjZs3%-~crT?2k;)Y=wl$z?Ydj0z}cb&#Lflox%d;J#oe0 zKv5Dj?!xMfPgmwDPU!{$)T$uF8edi&e+Pmep$pAxWkqPw!UO{E4QLd!C1Ho$Cb*Ir z{>)UtfiQVVYQm<0eqyW@QpVSzreToJSIua!PlKo;SG)ZW-j(!3S+Mf*9#NYMbB<_? zpVEDB+bbpr&gT-6Jz!*1rU8gR;Alk+74o)?atAbs){oPv$Z}|A(yZCAka8SVqQj69 zGfL)Zvq=-`1SUcMSgo?Hd_afk}k8xy-F+CJz`BJAVFHonuwWKF;7yKej@OoJWhD?dRa zxWriUa5j?sTskOT13Lq0Rs(&%G;V~5>_;T8Y}t-G3JrGg4)6q#Q)dw3HV>`owwD5C z`L9T1ncWxHb1ttEc>8Do(G1P^Qotl#>RUI*2oMF#p&ZCA_p#5oym*;b;Ub+gC~Blx z!8gSEF%gj@*3>5}jJDL~V<&<`XQh^0nnrGT&n$wYqo73i3_`2EiHmMN&CeNSoQi2B zF|QOFX)vlH9%89Y(CEDT$tTuKjR!okWkWB~dvb0}xoO*vBce%k%h6YjHF|Lp`MesJ z?ixu}@R)6A`)ptSwlDZ{KeZGx`ImQo61B>vVr61{aNmt4%86s;YU?`4>jrapLH)B& z&YPeABubpd5U_$LGILqz7{}otgCxy&rn%9wNS?g<#R$qc@1eGCzTv`h}rzy6rn*6Z-fosX8DAA zGdZhH6}fq@(cAlKVQt>1J4!1IkAz{_s@Zppdm6H%Y*FeR-&vc==JzhrtA5dUgQBJ} z<~kyiVskBQ1XKf zkkm#|Z&`q2F5c1%rOOc_EExH|BKQQZ&_IPTv<(DDPW4z)8#2Hbhi26oC7(rG@Ns}W znhbUmUfT}no2KoS$&AFqmD%p*gR@57y*d^&GJI4uAm!cu=WVA)^%GLdgZGNh+!^f~ zor-X7TJ{;^Wvbk2ldwcz+Gwuj*U}p-ajHod!MFlTHWOM|fS@lQ&CO-jwX>G(E?jiY zv{G2Rw?4yYqwsZzv{E_mZj)7qwi`JAkN~)F>u`6Km@jk_ef1s~Q5Ni#+Ko1i>zHZt z713biJrs*@|C)qYnp-LuN&^E%aPFph*jXLXe$hb#8v5TnmUV1( z^Cq)+8-23dny;e|wjJ}LS!}@o8lvJ! zm!y)4DsN_~8q4I>9_>nFeHulBBy!_po`lZIt0xm7po(Ck4h9yWj8w=*+%p=zG+QSQ znk4zpXwta;4(S%mB^~{F)adM>yUmck&212d6r~Urtp6!>Uz5BpWok2%`|S%-hRHtB+C}K|xz6NIRNfY%gPeS~+Rw*r%q?Sm&Pe4{@V z-8I`Z%6ZM&b(Txv?_9v5-x08uJW=F=CHpD64Q7k)@vOtR&KPSmW*Muoa=ivM%g|); z(v+~h7w{-H(^P0G72sErOe@8y5#%m`M>HutAJI=O_NJq{V6bHVn@y^+6M0FB#rFBmF z8R|J8Ll(SS!K6?@uA%%z$Q=7!tjMt=FGit(4>%3-Ce~w&PW1Jq-X$H|UFZPa~@xneAu{Q|JFh1?*d33_)Oio_6plfhq75&vDI zwT?Q#HE^y?4KqezxGFymvlI=}tsT0q%$r)lB|O+tP_7cxXYsC6*l?+ zRwW+wsnLuaglfy*{<#3Z#bN&lGMO++hdKmK{kV3ZCiCBaSxWLE4T0`}@4p12xw0%o z=k5&Rg8FenH$+EYF29UA=>UK*I6;l#sXYpD-G%P8PdzwDuO4^xZ4QS|$7lEebqNM4 zlgEpSIFKzfS#Hu_$4u}<+8#Fi$L{rVuG-qy&e$4T9Np}-$98Y9 zqt)k!uBU3BU$8bSanp}mY*%G%yCGaQCOsLp=2+R}^;aKU*9BDHbW-K>$F|9Zx7THr zX9qzqPtV9OM?>$-=JRuwmB#BPID&$5c5&gMvKSlbHKP78Y-1tcxYy zNg+|y5=<&~!ld!^%*=M22|rHA1K$;soihhAlD8taZri@y5}j&R7@K_mZBfLZVDKp!iTF@t@?hTRrALd{uhGQuvOd3 zBUJXYk9Mq@<9AN> zM?hnWc}if0kP*GEs|n!&*5Zwx_+174T>2E2=G5nZw#3m(?Ooj5VlfE=#sL<)qdXEH zN8`WTT(8&T0itT{D0!`{tO(K*WdTy~D-jWqG0+hA_}gnd7>o}$a4$O>i zw+51X1(ypy@d$qB00L7}Q`6AMhy|#O)7@o}YLVc7VzuLlR#5O7v5ipnV1t8gN1CMS zKi_XA0+E*f`MAk4agf3IhJCiHJPS?UIXVpDa3 zeh22cAS?@~S{ym@Wl~a7w%4frMvc&GFgVBq$q|B-zd8%cbZ-w9+kZ{6XpeS%K!Q1Re5k@ zh>e}pw0CoJGcE89mp031vS_Oyr7+d{euL4{3f2BKH1D-xO*=8V5#ez@KVQYQ=P!6l12H-+7ytkO diff --git a/docs/articles/07-glm.html b/docs/articles/07-glm.html index d208273d..9a04cd64 100644 --- a/docs/articles/07-glm.html +++ b/docs/articles/07-glm.html @@ -19,6 +19,8 @@ + +
                    +

                    Generalized Linear Models (GLM) including logistic regression with disk.frame

                    - Source: vignettes/07-glm.Rmd + Source: vignettes/07-glm.Rmd
                    @@ -126,32 +122,32 @@

                    Generalized Linear Models (GLM) including logistic regression 
                    -suppressPackageStartupMessages(library(disk.frame))
+suppressPackageStartupMessages(library(disk.frame))
 
-if(interactive()) {
+if(interactive()) {
   setup_disk.frame() 
 } else {
   # only use 1 work to pass CRAN check
   setup_disk.frame(1)
 }
 #> The number of workers available for disk.frame is 1
                    -
                    -

                    -GLMs

                    -
                    -

                    -Prerequisites

                    -

                    In this article, we will assume you are familiar with Generalized Linear Models (GLMs). You are also expected to have basic working knowledge of {disk.frame}, see this {disk.frame} Quick Start.

                    +
                    +

                    GLMs +

                    +
                    +

                    Prerequisites +

                    +

                    In this article, we will assume you are familiar with Generalized Linear Models (GLMs). You are also expected to have basic working knowledge of {disk.frame}, see this {disk.frame} Quick Start.

                    -
                    -

                    -Introduction

                    +
                    +

                    Introduction +

                    One can fit a GLM using the glm function. For example,

                    -m = glm(dist ~ speed, data = cars)
                    +m = glm(dist ~ speed, data = cars)

                    would fit a linear model on the data cars with dist as the target and speed as the explanatory variable. You can inspect the results of the model fit using

                    -summary(m)
+summary(m)
 #> 
 #> Call:
 #> glm(formula = dist ~ speed, data = cars)
@@ -174,14 +170,14 @@ 
                    
 #> AIC: 419.16
 #> 
 #> Number of Fisher Scoring iterations: 2
                    -

                    or if you have broom installed

                    +

                    or if you have broom installed

                    -broom::tidy(m)
-#> # A tibble: 2 x 5
+broom::tidy(m)
+#> # A tibble: 2 x 5
 #>   term        estimate std.error statistic  p.value
-#>   <chr>          <dbl>     <dbl>     <dbl>    <dbl>
-#> 1 (Intercept)   -17.6      6.76      -2.60 1.23e- 2
-#> 2 speed           3.93     0.416      9.46 1.49e-12
                    +#> <chr> <dbl> <dbl> <dbl> <dbl> +#> 1 (Intercept) -17.6 6.76 -2.60 1.23e- 2 +#> 2 speed 3.93 0.416 9.46 1.49e-12

                    With {disk.frame}, you can run GLM dfglm function, where the df stands for disk.frame of course!

                     cars.df = as.disk.frame(cars)
@@ -189,7 +185,7 @@ 
                    
 m = dfglm(dist ~ speed, cars.df)
 #> Loading required namespace: biglm
 
-summary(m)
+summary(m)
 #> Large data regression model: biglm::bigglm(formula, data = streaming_fn, ...)
 #> Sample size =  50 
 #>                 Coef     (95%     CI)     SE      p
@@ -197,35 +193,35 @@ 
                    
 #> speed         3.9324   3.1014  4.7634 0.4155 0.0000
 
 
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
 
 if((majorv == 3) & (minorv >= 6)) {
-  broom::tidy(m)
+  broom::tidy(m)
 } else {
   # broom doesn't work in version < R3.6 because biglm does not work
 }
 #> NULL

                    The syntax didn’t change at all! You are able to enjoy the benefits of disk.frame when dealing with larger-than-RAM data.

                    -
                    -

                    -Logistic regression

                    +
                    +

                    Logistic regression +

                    Logistic regression is one of the most commonly deployed machine learning (ML) models. It is often used to build binary classification models

                     iris.df = as.disk.frame(iris)
 
 # fit a logistic regression model to predict Speciess == "setosa" using all variables
-all_terms_except_species = setdiff(names(iris.df), "Species")
-formula_rhs = paste0(all_terms_except_species, collapse = "+")
+all_terms_except_species = setdiff(names(iris.df), "Species")
+formula_rhs = paste0(all_terms_except_species, collapse = "+")
 
-formula = as.formula(paste("Species == 'versicolor' ~ ", formula_rhs))
+formula = as.formula(paste("Species == 'versicolor' ~ ", formula_rhs))
 
-iris_model = dfglm(formula , data = iris.df, family=binomial())
+iris_model = dfglm(formula , data = iris.df, family=binomial())
 
 # iris_model = dfglm(Species == "setosa" ~ , data = iris.df, family=binomial())
 
-summary(iris_model)
+summary(iris_model)
 #> Large data regression model: biglm::bigglm(formula, data = streaming_fn, ...)
 #> Sample size =  150 
 #>                 Coef    (95%     CI)     SE      p
@@ -235,21 +231,21 @@ 
                    
 #> Petal.Length  1.3136 -0.0539  2.6812 0.6838 0.0547
 #> Petal.Width  -2.7783 -5.1246 -0.4321 1.1731 0.0179
 
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
 
 if((majorv == 3) & (minorv >= 6)) {
-  broom::tidy(iris_model)
+  broom::tidy(iris_model)
 } else {
   # broom doesn't work in version < R3.6 because biglm does not work
 }
 #> NULL
                    -

                    The arguments to the dfglm function are the same as the arguments to biglm::bigglm which are based on the glm function. Please check their documentations for other argument options.

                    +

                    The arguments to the dfglm function are the same as the arguments to biglm::bigglm which are based on the glm function. Please check their documentations for other argument options.

                    -
                    -

                    -Notes

                    -

                    disk.frame uses {biglm} and {speedglm} as the backend for GLMs. Unfortunately, neither package is managed on open-source platforms, so it’s more difficult to contribute to them by making bug fixes and submitting bug reports. So bugs are likely to persists. There is an active effort on disk.frame to look for alternatives. Example of avenues to explore include tighter integration with {keras}, h2o, or Julia’s OnlineStats.jl for model fit purposes.

                    +
                    +

                    Notes +

                    +

                    disk.frame uses {biglm} and {speedglm} as the backend for GLMs. Unfortunately, neither package is managed on open-source platforms, so it’s more difficult to contribute to them by making bug fixes and submitting bug reports. So bugs are likely to persists. There is an active effort on disk.frame to look for alternatives. Example of avenues to explore include tighter integration with keras, h2o, or Julia’s OnlineStats.jl for model fit purposes.

                    Another package for larger-than-RAM glm fitting, {bigFastlm}, has been taken off CRAN, it is managed on Github.

                    Currently, parallel processing of GLM fit are not possible with {disk.frame}.

                    @@ -267,11 +263,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -280,5 +278,7 @@

                    + + diff --git a/docs/articles/08-more-epic.html b/docs/articles/08-more-epic.html index 7e9e386d..cfff702c 100644 --- a/docs/articles/08-more-epic.html +++ b/docs/articles/08-more-epic.html @@ -19,6 +19,8 @@ + +
                    +

                    {disk.frame} can be more ‘epic’

                    -

                    ZJ

                    +

                    ZJ

                    - Source: vignettes/08-more-epic.Rmd + Source: vignettes/08-more-epic.Rmd
                    -
                    -

                    -Someone wrote a blog about {disk.frame} being “epic”

                    -

                    Mr Bruno Rodrigues had kindly written a nice blog post titled ‘{disk.frame} is epic’. In the post, he compared {disk.frame}’s performance vs Spark’s on doing a simple group-by operation. The task is done on a single-machine and hence is not the ideal use-case for Spark. However, in an earlier post, Bruno had shown how Spark can be used for this use case. This is symptomatic of what I call the “premature adoption of Spark syndrome”. It’s not hard to find examples where Spark is used when a more light-weight tool like Dask or {disk.frame} could be more efficient. There is no need to use a sledgehammer to crack a nut. In data science, what a “nut” can represent is getting larger and larger. Ten years ago, a 1GB (in CSV) dataset can be a struggle for R and Python, as many machines back then were still 32bit and hence did not have more than 4GB of RAM in general. Nowadays, 16GB laptop are common place, so any file less than 4GB in size can be considered a “nut”. Furthermore, if you use a tool like {disk.frame} which keeps the data on-disk until you need to process it, then the “nut” you can crack is now substantially larger than 4GB. I would argue, the sledgehammer that is Spark should only be considered for much larger datasets. I don’t want to put a number to it, but the largest dataset I have dealt with is the Fannie Mae dataset which is about 200GB in size in CSV format.

                    +
                    +

                    Someone wrote a blog about {disk.frame} being “epic” +

                    +

                    Mr Bruno Rodrigues had kindly written a nice blog post titled ‘{disk.frame} is epic’. In the post, he compared {disk.frame}’s performance vs Spark’s on doing a simple group-by operation. The task is done on a single-machine and hence is not the ideal use-case for Spark. However, in an earlier post, Bruno had shown how Spark can be used for this use case. This is symptomatic of what I call the “premature adoption of Spark syndrome”. It’s not hard to find examples where Spark is used when a more light-weight tool like Dask or {disk.frame} could be more efficient. There is no need to use a sledgehammer to crack a nut. In data science, what a “nut” can represent is getting larger and larger. Ten years ago, a 1GB (in CSV) dataset can be a struggle for R and Python, as many machines back then were still 32bit and hence did not have more than 4GB of RAM in general. Nowadays, 16GB laptop are common place, so any file less than 4GB in size can be considered a “nut”. Furthermore, if you use a tool like {disk.frame} which keeps the data on-disk until you need to process it, then the “nut” you can crack is now substantially larger than 4GB. I would argue, the sledgehammer that is Spark should only be considered for much larger datasets. I don’t want to put a number to it, but the largest dataset I have dealt with is the Fannie Mae dataset which is about 200GB in size in CSV format.

                    -
                    -

                    -How to make {disk.frame} more ‘epic’?

                    -

                    If you follow this post by Bruno, you will end up with a 30GB CSV file. But this 30GB file was made by row-appending 303 files (size ranging from 70MB to 143MB) together. I will you a few ways to handle such data. As usual, let’s start by setting up {disk.frame}

                    +
                    +

                    How to make {disk.frame} more ‘epic’? +

                    +

                    If you follow this post by Bruno, you will end up with a 30GB CSV file. But this 30GB file was made by row-appending 303 files (size ranging from 70MB to 143MB) together. I will you a few ways to handle such data. As usual, let’s start by setting up {disk.frame}

                    -library(disk.frame)
+library(disk.frame)
 
 # this willl set disk.frame with multiple workers
 setup_disk.frame()
 
 # this will allow unlimited amount of data to be passed from worker to worker
-options(future.globals.maxSize = Inf)
                    -
                    -

                    -Load the individuals files

                    +options(future.globals.maxSize = Inf)
                    +
                    +

                    Load the individuals files +

                    The best way to use {disk.frame} is to load the individual files, as csv_to_disk.frame allows a vector of files to be used as input:

                     path_to_files  = "c:/data/AirOnTimeCSV"
-list.files(path_to_files, patter="air", full.names = TRUE) %>% head
+list.files(path_to_files, patter="air", full.names = TRUE) %>% head
 
 #> [1] "c:/data/AirOnTimeCSV/airOT198710.csv" "c:/data/AirOnTimeCSV/airOT198711.csv"
 #> [3] "c:/data/AirOnTimeCSV/airOT198712.csv" "c:/data/AirOnTimeCSV/airOT198801.csv"
@@ -158,15 +154,15 @@ 
                    
 
                    Error: Column `WHEELS_OFF` can't be converted from integer to character
                    
 
                    when running the below code
                    
 
                    -list_files = list.files(path_to_files, patter="air", full.names = TRUE)
-system.time(flights.df <- csv_to_disk.frame(list_files, co))
                    
+list_files = list.files(path_to_files, patter="air", full.names = TRUE)
+system.time(flights.df <- csv_to_disk.frame(list_files, co))

                    I need to come up with a framework to help with these types of task, but for now, you should use the colClasses argument whenever possible to set the column types. For example,

                    -list_files = list.files(path_to_files, patter="air", full.names = TRUE)
+list_files = list.files(path_to_files, patter="air", full.names = TRUE)
 
-system.time(a <- csv_to_disk.frame(
+system.time(a <- csv_to_disk.frame(
    list_files,
-   colClasses = list(character = c("WHEELS_OFF", "WHEELS_ON"))
+   colClasses = list(character = c("WHEELS_OFF", "WHEELS_ON"))
 ))

                    which yielded

                    @@ -188,9 +184,9 @@ 
                    
 #>    0.51    0.32  168.72

                    As you can see the the 303 files are converted to a disk.frame in about ~3mins, this much faster than Spark.

                    -
                    -

                    -Load one large-file (splitting)

                    +
                    +

                    Load one large-file (splitting) +

                    However, in real life you may be given one large CSV and told to work from there. In Linux or MacOS, we can use the split command to split up the file first, and apply the previous strategy. But on Windows, this isn’t trivial unless you have installed some nix tools (e.g. git bash).

                    Starting from {disk.frame} v0.1.1, we have something to help with this. When you call csv_to_disk.frame with the in_chunk_size argument, then the big file will be split into smaller files using {bigreadr} before reading. This happen automatically.

                    Splitting up a large file might take a few minutes, but once the file is split, we can use multiple cores to process the split-up files in parallel! The performance gain there might offset the overhead of splitting a file.

                    @@ -201,10 +197,10 @@

                    # read 10 millions at once in_chunk_size = 1e7 -system.time(csv_to_disk.frame( - file.path(path_to_data, "combined.csv"), +system.time(csv_to_disk.frame( + file.path(path_to_data, "combined.csv"), in_chunk_size = in_chunk_size, - colClasses = list(character = c("WHEELS_OFF","WHEELS_ON")) + colClasses = list(character = c("WHEELS_OFF","WHEELS_ON")) ))

                    which yielded

                    @@ -238,9 +234,9 @@ 
                    
 #>   75.79   47.89  216.42

                    As you can see, this takes longer than loading the individual files approach. I am working on a guess the in_chunk_size feature so that in_chunk_size="guess" will work in the future, so the user doesn’t have to input a manual in_chunk_size.

                    -
                    -

                    -Load one large-file (no splitting)

                    +
                    +

                    Load one large-file (no splitting) +

                    Split up the file will effectively double the amount of disk space needed. We have the original file and we need additional space equal to the size of the original file to store the split files. This is not always feasible and that’s why you can specify the LaF backend which was shown in Bruno’s post but repeated here for completeness. The problem with a LaF backend and reading one big file is that it can NOT be parallelized and hence it is much slower than the other approaches.

                     path_to_data = "c:/data/"
@@ -248,8 +244,8 @@ 
                    
 # read 10 millions at once
 in_chunk_size = 1e7
 
-system.time(flights.df <- csv_to_disk.frame(
-  paste0(path_to_data, "combined.csv"),
+system.time(flights.df <- csv_to_disk.frame(
+  paste0(path_to_data, "combined.csv"),
   in_chunk_size = in_chunk_size,
   backend = "LaF"))

                    there were multiple warnings

                    @@ -263,46 +259,46 @@

                    #> 591.32 26.94 599.08

                    As can be seen LaF works and can detect the column types automatically. However it is slower than the default backend which is data.table.

                    -
                    -

                    -Summary

                    +
                    +

                    Summary +

                    Here is a chart summarizing the performance of the various approaches for reading large CSVs on my computer

                    -
                    -

                    -Can {disk.frame} be even more “epic”?

                    +
                    +

                    Can {disk.frame} be even more “epic”? +

                    Well yes! We can actually speed up the group-by operation that Bruno did by using srckeep. The use of srckeep can’t be emphasized enough! It works by reading from disk only the columns needed for the analysis, and hence disk IO time is (drastically) reduced!

                    -tic = Sys.time()
+tic = Sys.time()
 
 # doing group-by in two-stages which is annoying; I am working on something better
-mean_dep_delay <- flights.df %>%
-  srckeep(c("YEAR", "MONTH", "DAY_OF_MONTH", "DEP_DELAY")) %>%
-  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
-  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE)) 
-(toc = Sys.time() - tic)
+mean_dep_delay <- flights.df %>%
+  srckeep(c("YEAR", "MONTH", "DAY_OF_MONTH", "DEP_DELAY")) %>%
+  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
+  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE)) 
+(toc = Sys.time() - tic)
 #> Time difference of 2.800005 secs

                    Compare the above the to timing without srckeep

                    -tic = Sys.time()
-mean_dep_delay <- flights.df %>%
-  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
-  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE)) 
-(toc = Sys.time() - tic)
+tic = Sys.time()
+mean_dep_delay <- flights.df %>%
+  group_by(YEAR, MONTH, DAY_OF_MONTH) %>%
+  summarise(mean_delay = mean(DEP_DELAY, na.rm = TRUE)) 
+(toc = Sys.time() - tic)
 #> Time difference of 15.62312 secs

                    On my computer, with srckeep the timing is only ~3 seconds and without it is 16 seconds!

                    -
                    -

                    -Conclusion

                    +
                    +

                    Conclusion +

                    So there you go! {disk.frame} can be even more “epic”! Here are the two main take-aways

                    1. Load CSV files as many individual files if possible to take advantage of multi-core parallelism
                    2. -srckeep is your friend! Disk IO is often the bottleneck in data manipulation, and you can reduce disk IO by specifying only columns that you will use with srckeep(c(columns1, columns2, ...)).
                      3. +srckeep is your friend! Disk IO is often the bottleneck in data manipulation, and you can reduce disk IO by specifying only columns that you will use with srckeep(c(columns1, columns2, ...)).
                    @@ -318,11 +314,13 @@

                    -

                    Developed by Dai ZJ.

                    +

                    +

                    Developed by Dai ZJ.

                    -

                    Site built with pkgdown 1.6.1.

                    +

                    +

                    Site built with pkgdown 2.0.2.

                    @@ -331,5 +329,7 @@

                    + + diff --git a/docs/articles/08-more-epic_files/figure-html/unnamed-chunk-2-1.png b/docs/articles/08-more-epic_files/figure-html/unnamed-chunk-2-1.png index 2a7fd4ee27a5f44969fda398c7568358f547a0de..75e8627ba45832457ce5acb4dda340be07ff6d5e 100644 GIT binary patch literal 39178 zcmeFa2T+t*yETe=4Cr7&L14sI6a)qk5D;(%6k7!oK|;d_A}RusGw3)f20)vnL=!<$ z3rJ2z5D*ZMoK!$^ra_wYuibv%eBN94*16~W=TzPL>$7SKLx=tD@I32TYdw2z9Mw?c zT*kAEgM)*UdO-C!2gf2O4vqzBR%Owj3OM z*U5kLd^?6sIXHgdpsN0+=@i=6?0j!sb(r!e@n*Y}H9zk=x^QcZ>Y|s6mP|g}EpPqu z?BNd;6rM-Ml$bV2v3cHyb|oJBvtX;R*vmz`c5CcktE0MT&12&?v}4vLMya}4-K?3H zDv{#3vdzyI4NmQuaPDc5w;zzJPLIu2PVWibMx|5ck#X()+Ai&|fc%alhIgJg`MG!h zQYZGmNgjN$oczU|WAn+=a9p~*dT2iR&1wHj+~nskj~C{UpBx*vep*C+zT5tv_=pMp zv)z@E$4;I+Y2|zD@&0|PZ(@!;uB6Y@Uvx^Va>h zH*hJsj^AD{s;3$xM*H%_Pvd4Gzw6GLc-@rWENIXE`0en$UpH&Wb@N`Ek8hu?I|6hHo}a6F^yJBt{Y=Ht_fpcV##sB_n&IX_4_S}tHi1Hc*`e#- z#;T6U#t2B?rBL7Blh~4&)f4;8ckY{S-I)-JN4hCzl9)C6g)f%~Y|5W%9{iG!-V$p& z^Y-mq3HzRp8Do`)+8$h4c~dm4POv%X%92$+C1Tk-qYmF+9jC)ceC~5dJa<>K8Xh6u zR$owXb|fhOiG6?WL<_GbH^+;p#B(OHmu{}(=C0MJ_0Br$rJPaMv$B|zGe6K;RDE6Hk|T&N7xcWr_Wb+fAQL4Ttwe0$H>P+4A-Jtqy+t*nZ78f5E z6SLh%z9+_Mq)5otaPG@3apS(aBngMUy1G->`MyL38OEk1GD}0{q{{Y;zIXb1-%Qv* z*{m$osV}MUOHv{KEiW(a1iiQRwKVS6*EnxYPfs80Ri{7MpMT}%ty?BoDGBqM_V7^he z7TG&VtsYZv;~057TwJ31X}7ty>{n$N<{D$NtVc^_3o#R4aFTq&6!Exw&a}$1Jioahc~Y zJULNcR9^1;dX;icjXej)1?|<6PY(r&QB^5icXNhbEthWiijf{kzb|ipenI-PuXzdm zESUzY{FzF3rsv#*CxQ%T^3$=7Iul>@so^q(!CelvJgUcz9cw7m2$R=3SzvwG{``pC zLHY$mo~xrTl1)qKuU@@+E|A`2bLhC9Uhs*ijEvor)eMh-!0zfeolMDIgNrX0oy%OX za`!KjUqh^ltLM~*ndaBKy*h*Q0KIpprG@o%S&S2d=&;VwnR@~Uc!HG&)oEA z5!TI*PN!3-boR$)Ts$?>mr}ls>_d&oDh-7MgRoV~(-*VyXxR=$!P4gEndg^;%do{Y+`_=|Voo!zConcND*mP*$e3=DAOuetUD>x*9D^>n6Jo zzn!ZOGIV)bEMsU>wDnv)4wat!yrId?2%-MQ^dsp4iY|6nJBki&-i?ex3V*TveZggG$bg9d&~aY2O^ZG`cgWIo18ntx<}q} z8*M5Lk!3tf#35_X8AM#v>z*Pr&|R0v#dSoHskvg)?q`)Fy^T0NZ4A#DbCz5CySq&J zmMLAcumCa7;kVpLf4uyYGj+xy0*quk8^N@Yyk{-aV8#{kHnAN*u0Msrz#>31qqj=TQeeP^5Aa}-e7(GcBHSeNa1NH0ggG>v>mC^{6o`=0zNpT*;n!bWm_6&b%!Mw6 zNSYo@=#6>5UjM?~Q8#8!wS6KEQUM;ztCEUVY##ie-_X%>xVvGu2it}EZZF5t1-oq<^XWY9>rdGu@P zEnce}iIsTDy7O@fz$NRG@K}Wsg`3vMwz}D-&M-9UXHuMozl<>hMfIZ?M%HiFt+UiO zFxYft&#>zB7w&v7%=!)z-}Nn0wKh4o-?i>dNRQHar}@?Yga)uC-j;jLj^5kjY~yX_%`i8Tc`m?N5!4XttK^@0w&GDnXSnN^H*1dE9&{gctpE-? zTOF%4)>kj(f#om2VhM8|tdT`f%jk)XGmdZ{zfqH*Ux=0EQrl#$+Jqc0trT_M*5hfp zNoJl;TSIL5j8Q=Hlm$M8i%SAixx4MQlrfH*FF|bO;fihfv%UFWyk%SJV%%f8C8TlM zx9{Bf{{F}FWubE2X+QyTg*5w!$Yg>QM%s%={(S4BFtDLCSo#*Ami*#X{F@w`?R6si zm1n=TRZTbz=1w3St`pIIVw&FK(kon|9j&A!%V{kpvVFVP9E&k`bLpCGZxjY|PFPGP z>%1CBDOZx~MII-~l`Codd~44(4fDz;R1%`PNAF9S1&T=Fft^_0(N&!~G<&U7i}?3A z+&pUh&ZnC^=Vp>HXEmAOOi`n}Jv7Uf9A~?pk59|&6)~Ld)>C{64!5zYj2e4_7ZB|E z@>$i4FBNJ6-q?cyB2^DnuG={`KRGLHKRsMnh+Xg{Vs6?j)P20Zqh>If7jt#GcwE-%nms=mxAr zE%f0lu~0ty;lbkmL+)dbH?xv8)C*eT8A_C*h9XPAl4KtDfnZ>`jwBsDJv|E=AjP2q zUx5fEMlZ#j(|TuZt+fF!Gjq=TQJ;s1uyE{nsFc&tr}I-MjoAAG3F&-mbl;O>(qX6j zGdsgeHl9d*r1slyzl{vNUKP}sCS&q7kWV24K!07}Q^4Knrd=O1$7L%-SfodVi|v2>4`1AxlY7JNp{V)!!fn25{*v|ACIY3pAAtHaNpJZ^{+J= zF|`P;7m9-yYkndKNO^9uoQumEkhtKtC46h+v=eSxw-vk{>#o{j+^_X3qK8@2#n+s} zO}6FxuCL_#0%+hJ=GY>qgm|3M9Tid;w;K6+=$^8thYRl+8iC5)7Ote8P+6t8OE2A$ z??kkQHwi-6#%l=P_n2Y54VAMkZWhZdTf}YfvyPFwJ z-e~1}M0sYHLyL1K!SlivzpR)Q?=oIU3pRX6LIkM+&-Z-fj7jyEgC8^@=E$%}99Vd*xyt#o> zeK3Yh6Pe5|#Pv*;h9dK0Xrt4T6&rp=xp?9E`D$TO%VtasWSgKI8jhHoj9{iI3-)xL zOk}iW$n?0*)z$!;2*|cLnW(Cg=x^O*UAPZJY(vgF(mF~C9XzUtp=*EY*_!zJ{kCas+v`k&XvVYC zlh%OOZz(;^;ALY2BTO5N~@i{ahD&x1pCyNFg)U85j{Efc3LmyMFyw ze^J-QIr5b})%m3|jZaDtpmj5uzDlgKSq9^sE%)e@8vwC|s06r363a4jz>*6Wu1=)p z`YO*lvD}cdc$;d9cUiUU+3(F0&?0C1UZU^~w#9lZjV1O${wv!97MtA1Dy#u})l}bX z@!hl9$aBWNY>0o|j;VT4g+vJsj_R}T?e1rpcZoP{5v;c^9XUPHboN`Ja!nR(U23=A5v3hEq`;vAphtEHXFZap(%UihM@rf| zJB5xYO`gt_LcEKU@-SQ@o?2*gyAR0+9TD;F?Uy<=kS2S5Er?V+zMM4pP zj2%HnhM)OXn=<%!Zr^?&aHlRkvGJbHP|ZYV)aLS%UVWhWOq1<`bU^8CDQ4w~7ela3 z=cpwJ6wcBv5y?7>A`s^~YSJ&%OVZ|$#xER=h&?ghlrfN!GHcR2IL8#YzRBj@;<@lS z#C_WyHw3+&r)KL!cI=>fjHg6s)kPpxiEjk-9w`=KP{_iD^(d2ihJ_eyWLht0`xR*w zTui9Q)I5%uU{ne+`(I39;Myn4rzb~ki<#Y3(Ur=tP~rv$#`TToG%U;sNZmoTQHkJJ zbn(|W0{HKH_Il;H`>~dnY1qlhqh$&@JS$2;thmqMOXHu1qiFg3`nnL7jp{Lc%Se;8 z%358E`(|KZ;8!;p;pNq}y2{%p5$N_-b?(EO5$-Yl5?V&{2JdWEIlt+bS>Ub%B`r>E-rc>tQlAnA z%hgrFgK8&GwP|md#^Vb)cRjo=JpHPt#dT1!iZk7l`6|rjwSw_{oncMiRpT$xZN1}Oce3+3Y9uFQh^kHmJ?nrDanv|n4XNfNgz@fDJxY@v z>x_Ch;c9JdZD3ci&l)G9;GTrZmrDy)dU>U;u^}^=RIW7TAF|f6x&0c~vHpv4z5e|w z^{tF`idF!7smx}V-h>LUh-X2fAVjHoczEo>{(7aRm7i~J0=#BTVko!=gU`RNkxsR= zNwt?Lu&zdo+5P#iHH3YWFfH~cl+@L!Nw9C;QssN3NRq5W@-yEhp`Y(DQ)x0d=sD;5 zK1?C~l?~$Iit?kaDhuXuTY} zUPM&1)Oi-=e^g2JRSPbznW+iJm_S_e++Rn+_sFf>GxXvf=##bU)@kWmsR%0^Vqy__ zoSmx$z6gb0{5{JJnO#fI63C(9SwZ#hD&t#>+r}dErj8yx`m;&F?Hf-|L>C}g9qqpt zX?i7qsT?GpKW&rc2B5B`Ukh5)z@pLIM8&6rwa3T&F#;nkfYVB3G$mqYC(WbhX>@Vw z(}%2QNoA|7a+%s$ZtpfPyIeU7Oia2^);aCLqy26mO+pr z`UVd5rq%>7cSlF?V|^~o^6ap$O`&bQ(pnIwUGBx#g4IL5YOG3{9R)J3 zj>)t4#i_FKT=h{CoZ%K-*L3|%wR7jrVFKRv)u*)KxKRu5Mo61$6ON&K)56r$*^G4Q z=IhLdI1%Hmo^x!1tJQ5(9nn^3_{D-%Z^{uqL-RYq;7*o%?nlAbh9r}aT$YoQ6DlrV z;_=&wnh>kJ`yNl(oHq7f`sLht9CbNAIXyjUzvi(&=V9P0{d1YYf9XkA1MmyiGUg0> zjRH={eAJVz(9sH(vEa#dok+Y$19JTG=W;P{4--wcNq`n=_#0!;Xk{?;gpYY0L*GO`c++ppF@fC}IW2#BSqB7U1HI7U#qh1q22b{JCiP7Zl64ajf%- zZ#WYGW0$$D@b{_lesj02hUypIB^g8%TD!-V&NAYH#~$~YxuOtDjrCMvFPrJ$!g|M6 zciCcBnJ1hxO&8vej>rS%ADZsB@;nrA1mt3;o6fk^pP*!_W6F|s<`J}GMoT$UYE!lp z)jq>Mq&GG;|5Wg415c!;6D9rA7Sc@&~Xigc%+oM^2 zt+=G*7Kt+eUc3k}5hmcjhf!7$qB&?(N(q?L7|gL&4XLZ5O9B?XLOE!j{kQ5%jIz<&81;&XkA%(nm8UamhV0aRTT{TU~(NXg0 zHCDmVscyfSmfSgjx!i{xd$qF*V5LA_*`F`%j}VKr`t5_dMm%3^QzXhuOzj6-6iNDg zLvw6i$HmoF=<5rLTR~9J>{~#wWucK#eH=|hUWkYd=VJB%+MgekdYjt;Kj9Lq=Ek4$gbKrJgyIJX6jM9bpQI+~x z=>pdp`>AUz_V#Cae0wbrRl>+)ozH4Q(PB3VQ2X>4&B$ZYG%E+>YkR5XUSCm~ zKIh~_Da+9uM3sVlY^T>2p94V=p0ktk>75{Ry@*~0TI^1WudF6Ad0XDpz@h6FMB2y! z?WScp+G{i?Pvgh(lx3lOm&t8A2TqZ<|J6K>OX@fBAJkp&&kab8N_VPhkErtDQ(!A2 zPW`F1#?bRfwMp`HN%J)&%chJB)cfbFzc%POx}H@nWrXSxd^B2+B^V#C{tz_Lb=z`9 zS)!$#%(O98RYk>7Fp)nuLpZfI$WXrIVC3P%5WTDo6=LF;dzM}I@!m9xcTav}K7HuZ z797KvG18IV=*L8`(Fjrib`#rvu21PgOtQ{vPXHh0aWnh26w9s+F}kgt0Xr1V(dx_3 z)JIf+T{d`pDW~twDiAlenM{>yE2u#cD1uKWpQa`>W=pJHySDwdRyHUs`D$BKW|1X8 zC<&uMD=;Yc+q|-3v!lv$DNP@K8qA;b00!Qigm77N?)Q{G|NJx5Auu=lks#s5T_y@R zWt;BUHqmTLrMqqFKW~E8DmPCYbyvFYe$`Z+VAc4U==^O`szIp(%MwHqGXzn$`F$HH z5w}g;!vdAi`ap+o05J(~mf(szE;HQ&)StQ3g6DPA2Vb1l_xGjRFl+8C!qs~Loe(Z= zdOSVfn^e)Pv2MR>%eAgvCWJsjL)!W3*!{dqY}|R%22lc*Aq*hk7GK!V3RBqvd*Hrp0Xw@l>eB)%t=W$03x=R|gE zb&NN!K9x!>b9!Z}GSsl)*FX01Toe1Q2I=+6D}nn$MK^m_X|6-u^L|Pm!sNCi;hP}L zI6#nns1cEy+^8x0`4os7`D5)+T`Eu6t`{@dKw^QH7og6RG{4MG9G48{8_MvaW;u}M z)*0rgt?vP>DVKgFh)I&}4WaN2*#lYT0qKek15I+B+HatV$aX0YnJx&^x6F*(FLfQs zdIl>I_k^d4J^d>6jdaACvEAc!CUH5{-w-Ae>}d7FB7+%iKIa>?9gI^7w2|yAK6c`e z7oUj`?*)3b&2j^TYmc9j89Swd{1nEY6Z%gtfMxdxA>6g5Qo=C9L?0#(>bqy;04tJO zKYW{8+|a<3kDz^9*OcV)TeohtMTJ=2B=vwLGzA{@pY)VE2^VT~aNY>S_dfU4v*2PfsW!FIiV>=|q+vLC!px zdS1&r{Ro|MjgvP}sTJq7#JjB#cp`LZ^L}4`TAWNGL=ZEZ)a1>crp0O9fI66Nb`fbL zqTnpb!I8SNShj22Z8gp z)fY?wyw`vV(`X#Uo=YkoF}0czYm7)%arfuESEB=40#Ovh8h9$cP31(njQ8Ew(95_` zpDbLyl#7e-9z4&!c6Jt`YQRqIvl!<#xcFP7@*uQJU}lX7hVr>{x~_dmYPWf$h251d zetMzjdj!QsrOi}EUc5o3WOQq;3p19v(A?B}gM=WThS_*u{qSpn*=Get+B6gu!|$ca zpOg!Lv#~K985yzh=p>&+Yy>=4r$n^ly`ZW*F$LVxw>%kp(++u?EuC;-M)irHNcM@B z7h&S%f8)av31Nxu1_cGJN#?_6u50;{jn|VLOBB(78Z8;OY!Z?YgyNnxLEAI_HuSlB z6eYE6SNmpXXJ>*~pOjzcr|14MKRh+sj$`_;kf4Lzf~6TMk2bM{FP0lPr3 zUNtF))8e|ERQcTJRi5uQF5mTNcw7bA3F1sCY;qfIy&C+IX2x@5+L9Lg~GZ(lxk{i z;02C=y({+=b9D6p8{Gya_^|jhm_&;`n<0l$A0k}Ct+RH+qF}>1biQ(pa z`*Q0TY9eTTkY~mQniXbeC-QaVKURSlX~WL6*0|^jW^bP_zoPI?W#FPv5i8G`UL!SF z4%$#3-Ud;(-OS({q3B2<0-1x2hVg~|iC`S3&NnfZi|72BfZq-r@?~Jv2KD7$igWBe zLpgmMY`&k@0YSn0G8TH?X>t&Ot1O_+>siiVZfR}uu47$y18GQ?ksVNz0t0&iwGtY& zLS(j5>4Dx}kcpqXl8g?uIjF%Fw=)5!{D#{Kg|?{q943pQs#*yk7-|D)K#oa3mt3#w z#6Xw<5@pf1;Wj_e*;@%oSV<7i44e}8cj~9bHz;XFf{Z`sjw&Vgc@w~fZ9@%2>~>_# zBLvtqK|m`aFd~#ma$YhVK{m`#r!P0o-6ve@l?KD0ypKA{JjvpOsMQ-p{;&GO{bbq& z-Ac$)&kc}4h>3%cBvhg!7$8~{fmEnsf?6)BFDwZX7uV?ouC#5*MrJHR1o6KupsbYn z>Cfer6VQYX5R%S|$m^P)Jz~m7579L}*($^d&Lp-9VP2nXn&k#x3I!YV&;aHh?8#<1 z_dPu^j|ro=%SoWe=5Rkqt!^0-h4^tO+YwBt@UB!zR=Wp6g)i{o3#sbFewD}sWlq1& z@3K8}xPaFWOy1XcPuP%(t1B2r`Js}^-(IewV+Oau(qd$T@=^#A80-%09B9;#Ze?`J zRajxwt~YE-@b&do<@G`+*mNY%yua7}@H+oNXtP@-Fsh-U;cBjKiS=AXoe`df zwX}sCd+C(CF+IFC=v$peTb~MWoAd{DPUj%r4!>G1t|@PQSY;B~Vgtza+hplHM*JdF5#Y9ni;IIi;U?CmCaeOx z?t?L-nAzWW`b3wSJM*Cq!%)I=c6x(?!-Y*ah!v1m-+?J0rOzslp)16lMHp7=HgDPe zPz}DsJoY)5&WFFfYR``(pF~7(&`8FgUR7KcPZ!bAeJX ztB~E#(`u1*j0c)Bhimg^r8_fv>XJ-0LEe?H{_^G zXj!V@8OORRwv@^D#ycTekPOoFsj-09HKKOE-!^sQH(p%qry-iw(5IJdat!C{0_Za~ zBMC)pEsRCmWp5bmBLX5h&M>`+kcypD@8B{E5ym$XWuH{P6xWuy@I4B5xCc(-@?p_X!QTtz@r2 z&haJ|H;h^)V>X#cjRcwO)J>-F13VCFv=2gT1t=4;mgD`6g+xRRGV*YK58EMOthzvD zd-qx)QH9#q=q9;MjyOS%ZhLe~?B*7q-GZ<)@em^&Tt;m3C+63BhFM!kN+Cj50(9kS z%*X*o*oHHB(@;ETnbRo&E7})@WMWqnoo|b=#j|ACbM{+u9o5%W_9;hPBQ}pMg2wn( z!e8TG;HzzwNhA<@;NTlPUSuV;qBvN}Oe1R#Hu<#+JV#Z0Oo;im`gsyB2k#+(zP3`? z*4yxVNFjR(+*&UhJ5tlwI{j6)dj!6eHt!162Rz^3(y(FD$eiwnjLk(WQa9op>!Z}2Zc5>EB;?}Md998ec;7ZnAk4#ws! zRVZw2>r~5xG2YKMeqLFnv;nG;M*4je8PkpVGx{dd8&8&hwcEsd1s38&3)}G`|1BvKJ9-z%m0qm`@j4V=NsxcW2~t3H_$?i zLCg(<`&R|*BL-QbViS{-aMKxh07+=O2lbR3aFk;k0f){Kj{zv)<5rFAr~VjCZ13L9 zUaVvkHyK@zZhe8uc>~MwmTW8RQy*xNFpNsq8mxgf`@a4R_+Jvb2sFHqQYu3nY)BEZ z#;BVlU`W~B8=q3VD9ye;N(Rd6YAl|83;)?oWT}W5O!W=2A0Jqc3+?OY-C2jFNqW0p zUsY8VbS={JbF%h<&&_Maf@Nr~oUN~}E;m_o)8&s8Vge&cA}p=>N#dtm9FI0K(AO&! zcA^MhKRG!G&JFKdjQ^$v!7Y$;W1qF?-^Et!}%Rn|WR8<9CpZbbPt1Jy4G~%#}}r5>1R*Cg5EvfSS7?^X5PkAsdX?&yA5F zlR(%+HHsSM>=qOZJ7?og(twv29?s%}OfMWpH<1ILRWb>rxR6?{9gw&rFRVD68+c}m zxRqZfb^fKr+74bxP;$ns4syiYNS zle};EVOW#EcM&m+H|<1&`Z{%v2df8eCuFJ%eiG%4 zT8YyBm1OJO?1TxdYr`O9@YX&q_26AMNUnvK_%Wj2E0KQK8|U;E=iB8;XK{(CeGdi>y3*zwm&t(+MaTae9G!v@%Me-)-c| zJY_)xI>4{9KZP8AkxleB)igtM{@2k7~4 ze{bS#pWsrmvcv8U2aQcgV3a9|t`J0xK>}BxL?#%F#Ukt&rui0e@1v7|o@a>NMq-Dm z>Iwm+JqVe}e}CvUYCUCF8{ttQ1vu=)mPf%T0Or6o2Ov-aJsD^5QKaG?4jXh6G7DL> zeEl2LJL@r#NjOSx$(Iq?f=HpL0oI`Q-^pE5V;a&xcy=VHkIt?Stg-zC`ILu|DrH2Z zIHUtXMQK#RmKrK$weZ75;L)9a@$=jc5myQc!UdKE`>TG4lJi@h7UgXzCOkedW zW@=cywpQ?{guY?e%}*o!38We34GLBPQU#Fq2-MUYVI4J!L}yRh(;f4~Py1cszLd3k z==ZjthvvJvE_E`u^jq#EyJ)`QyKQha+i(4JwP5q7Mc>9(57jhnlUmGfnR@Yc|Nm&K z)PFgD{42hg4d*fbFX|(U0IFCOGDl6x6=`YwDNT4yT z3@8|}UfS54gM-Hi1&vHx{*9559uL06OB4DrKmfdBDvNjuS#zNZKJ#L3f!$R7p5!Ss$bX-*f1k4d)>HQX)9EC&F;4@@#qjF++!D*$ z4*k@3SxG$9L_**Y*%o+<-Oxe01!rfb)IjF+Sn#qB9qT!DNe=O(^IB$X1}iQingW7` z^iYxlD6RfX2(h}MR%`JB=f8zmxXr@swKHF&pL7WHp-(%QZV6IH+>D z+A{uRWsRi&F7JhAf1J*^pWQ0J-zikGV*cye&#t9y);^u@4O+`PgVVQn20M>**aG9d zNYqZcoc{Rn7FeY_WZEp%)YQ(Dhlim(xC2yN8O-H$N{FmgSy0|s2XpU8D3&Q#!3Izb>|P0>uH3(40P ztp@nD_TiWLR<5qDiKrrT2J9InV5m|`Q!bLe4@wbqi^dM7%1Sv3H5sV%?=O2B_EG_a zDG@RW|9k%Y`LxV&=wpZJ&;y^8^giD2eUOe~D@)CFmN7m)zDritYBWps+!dR+*w~S& z4VU`IJ&mMtu5oauJ$l^9uh4lnUyO}^x+R2F4d?q zran%nii(OVboDl$g3O@DM7ELrj7aOkgluI_C%G?q)5~+DT-h^o z^SFfLPz&1bCQFQF>1e&{Ww`8mJ$~N8LVcu?wl>@52jlHR{{MGF3 z?PKGNShgv6=I80@6C*C0(FnVJ0S~JZo;AeX>4rq(C61xA3ofHm-%B&jb-ZB4qF?w+Z&RqT(gLZ z<<_g;+B5EAVq(I>%bVWYx}eVeV|BGXf60fYrciHQ*>N(6eHdBigfGURo9$pBeqBcxptsD$TW5x@io0#DuC6BPU<(R8A!>m) zZ_u@C*RY(98_kL1h*AXOtGqzteMR=1?Thhx;re3U64Gu;Gt`F{Xm+|&xzlR)t5q@z zB{O*I!fJ`8LmE-x!Yl)kkk*imgH6#dXb7ukUsQ1PubV&Rz*g)=%2Y8%GbSPvoT(J- zW;x6RtL85aXdJtnY1?_`<)4e>Kr*Liq=6tIbn`~Z(+A*LXny?9;SBW8bE$&Tn1bGD zQ@CbzpPxO}*@kJj4^IsJH2PjJ!95lHWb~H=HQ(^%C=O?qyNi+5(L~5@Zk-CPM5!D_ zyWW$`OgS_yzef$?SoHlJw1n~{Ehy8^pz(~oa8Sul!z5J?dS42%&nC1&&{}R}|6@xz zq{iaN`n2%>DJm@^6NO!qUdZ8alt!cV^!EBu#Phxx%UCwmm8_X}bUo&MBb;^gCy|j` z;VbL!xbeV>P+FBWH8qVU93QTtQo$gw^3_XUIS z&fY@{NQZtaOS{vgzrX*!ob4G{1@9A*4}HMzK>H*E*9xk!<9E#NtBYEUP`#2hC+a6r zs=|EHOu;0gdiqIMSeeeC$nWmz+I}7eGi=wJCk}kFIDftvoJ|N~1Eq+#FWMus9U)GP zP>S?hMgsT1^M!6i4SM?$}mMoibyXMY2T#MTW819 z&z(PikWN%6&|DWG5#_j{(L-~cru>9-Rs z_&yy6t}bi=@4c0ZHksPVIx?H2OO9%T)MlBKhfz}{!O#Bq@+&Llw^K)mnP)V~2+cxa z0;I7;Ki(5g5BVY>gt`rkA0Qo5#lq^VQYiu<(4M*)^P0IC%kiLTM_!=f{4!8 zY?zOKm-j`Hgw3BjK%DL!Qo)&>vJ|HAwm7np_0Vlt<=hjiljpk ztH@g(9MAkmba~=pw`34z++fU}7^4);`ea!2cErE;0^u4MJNy)sE$pcX7*xpvPtRP=gczkD%{%{1RvTAe1sD#Pw*PDY$ zdK;Cr3jDqB_y9~5g(`*v4LD(CYhi9~PUdDCl_&r|z7D5+8`T4r_RicvcPpg!EQ53u z-k~0>ZfB6%bPDlD-rDdTX9_Y!zED@mcRiS{Ql2?M@*Gjwp;wl`9(fP#d6FbT!3CCl zx`@69cB(Mfyzx|qDfCy-ANI039k-3h69*0%Af*WUMhg1a6KbkIemo6v+PU1d<;2JB zWap58vxSHLBsNwUJ`-KCe?6ztJxhQA!dWa9r3iYqBoZcZD>%am@rFj;bSdm;9bb+( zFZxPJ=f43ipV{ZRgZ(c`kg($_G&Eqd4#%3KMRSPf@Ivy!qaIhA zUp7`&(n+TSh;AxCn!WwO?D zM#uVl?)?QT3m`xd=PM_#;@B&J4?#1oIvX^5MYEUbmqm+Oi1&sn2@+N?{5#@UyVmK5 zeD{-3yGm*%x)YNz--3DUf$WW_Wq1i$oIU#vo&aVjx(?-llO@q6L=2{|@*hF%5E~nt z)$9si(prht>@R-1gEHj9uP6_A92OxC_7^NjOJc5g{%`CujH6Ne9T!`W;aJPKqu>-0 z9DLJBk3D9?hj(5mU?v5_*ONbA(s3&FJ$tc0jQrt6;xVUBWGJMsA|3a@JViV~eCp}x zk;Dy0Zh+KfkpFeYej=~N=J@c!6)de8?5f2id=PI7Mnt6pzn+2CGx^?3$q!o1S9HjT z+N>b&xpY(W-NqcM4cs&Bu%!1-Ps1lsLUJKah7_^LjS?_6{N*i`e#Gs|nvIallt%9Q zu9aR%-oBSSxQvX7?#73G=3$<&Zr|1;qV&#YtGue zu916qFrgcdAFMon6cXAn1!cC>q2xwXY5vMtsWXOs?l zZ+1tXFftnS{9}72#B~yBa}^hJZ2brs4`M+M5GqHP7O}EMu)<>^BTb_x)~|L#uVgs7 zr8{8+FG4@M6!w8-E$%UpBOP<}cD_L;HLf=QGN_qNr7N~->g$J~Z*LowqyxAS*mV&p zNuaufQqG~GvbDASUX+kTfNGgJzJYPyf?1^ljDz*G9ldeHO%OGm!R)}`tQgEsKKb03|DQ1vs{(r==~+;kb~l!(Z{CqT;jk zLXNIcTqD3jQHXj5#(5H$&qlJa|2JDj3UPqC7)Z=xefV zwy!nCD}Ka6)Dd`(c1uf3PkvgWe2(SVza9 zP7#2KcAy=9pQe+a+|l%e$-HL(122p^u4!3$a6A(y1~ER`phFe2Aid9GK7f1F_02VLo>@p4jIrDu*4q@krl3!qG*1lRrHE2ctP zr?*jU5aTY?vd}~$M@@!r>|ePuw@8l;z;~fGYuj2IlRA?FYp%iTmGN1z?S|nPWJ69B~#C%FA%5GsM{_9H?wvfh$r-_MW zJ=JlxaK&}v&$OUkf^kDHzFO-1<%ZDl6DRyu0U!STl&p2bPhiw4h1hLFKo86ba!5x~ zHtp{YhwhdjmQ56_egtVw0&zEFnjy)Og5ie&TG2_gC&!fpEv$dC@SLGnDV z*V8-pn7wHN7CA~ylbHeFOfpla&KLQo-95lPe|m3P0VMRs&YcLO#esE!DnNQ=Y- zkT$sTB$z^L>0sZmxVILE1zDMaLWIxmgF}L+b2Cb=V?S(Lee4FxIdk*IL}qDeDU}XX z^y7BI6&k>I%s3;9S{8f=Td9N({P6`aW(ZMp;KT zJL0&-?lOf*aUXH40qNac#+d{Ds*9X4IdqWRn#HA^_~m}^l-e%#?~d;LC->nn$;FA_ zbpX^$P%U^fDX^L3Vmh!cVP-?57c1N4A21{R(0nVvZcErB*ti9DbkacQ5+@A{VkrR} z$mV5*189i&2;7UXC5QWp9NvD%z3%&c7YxFKD2eXwvY2cPKcQw1t_ipg!-ZK>Gb}8} zjf^5-o!AL5L5MbZ6haANQsny{l*+q?H|8My^xV!y=H``x^*W&@6rdZyn@T67moxPL z5fl$!3899$3ZW2BqCzeEk%xL*);f`9GHclWX)yQ7>Av2bpyeWoj@b6P_9p zrW_4<(H1xWD-lIJtUMd&GW>hN{~>y8`w-aDR&_d{^RMJg#Kxt2Wa0Y3{L!{m?p-SS zG|#G#7JpF8TdAP11(<(Vcxsm?VCyegK9}=pg1=+wgdvxI85y!w|USHhwsT0`c8F>?ND6C<}c~( z@J5v5o(2$<=sGUDICa<8?MY)N>9`6MqXkn)-gEt7{8NdJxgykLG_VruC7zf{*a-t% zG85LTq0s+aY{C&t1v+HAr-?!^_~UyOKsk_iQEkABo8efJ>dTLG?-GwEy&azG5;Pox zZN*U~y~LzhfPNaiarGcr|Mf|X=pz@=>PakjgxDA*24iv?TxTi_;+u_dgIIdQC~WIW zXhQ0mo+fFS+{9INjJ@4h{`}1^Lnyc{pd2mN1Oat`%rArA><+obu@%9{2E2AumP@a= z1jy?$-*Vs!5FYvUUqha;&+)TV{#JYhDl9ozx)fUN`m3kglxI!o?IaPAD>g*F^K4zV z4%AQ0#{`gJT-!?6phVN3U!a0LkiCOH|C{&3OkAS()HZ^$>x&2Oj!sHSN-w;`9|v55 zm_(<52nW~K18%t!8cs4|^7!}$j-$h%h95i+M_T~_#a+-*x?T}9@OGWf82jHheV^o` z$LD{)QJA_khdt|Cw|=+b`ul&k%lyB;Kr_FE{|_BkAkk4&VB+L{vr1?jsK3mh9gqMf z1%6L#0=?gx(ePgmzh7lt-96ddZgn&hpCI8>Q%kEq`Fo|FeH;oBLXoo3R6{MDf<40S z6Cz=i;Kl*m$zTBZv{AvKZ@nwX5lPUt@%rgv=vFBuwt6%an4)7xZXl!0reYtr#kvcQ zXV{Dv^gM~n6Q~NxYr%kgtgY=NbSErdVH8A^cTgaM2(b`>0+pP*=%}cu(FXKkrQp1i zb4%{~p`Qk`GFcgv=a0nQ0U@E!&CP6rJzr}d8kc>VL2o;-OP88*1BysOkq{p|QAr@R zOrxym${EMiUhDy)eJJgxWRxwLxFpz*LT((DmzQ^W{ykUxv~GXi3gN$)(XBFPP!OAuGtO8 z;y-`>+-{?1NJxku1%zB(w<2vLS`Kl)RAWCAs#8=_(s1?MzRS(UNUHL4S>I4@6HUP| z{3qd7%%iL8LXND8mX?7|9u5b+6DLkYxbSrs4co5ccGB0=8$rLL5<1<=(H7y4ru0XR zjhTi2kNTd=^Uk@RT80t1b<&a{DIY;3lOAy(!HdOsIiE^^!FM z@4s_gf|`d3(eaL`w0qFtMR+XS_ZoCz*k_z3huf#=PfBF>G-gU@Qs3J4|NZ@MHs%03 zu)FPY>{MTo+Rn0;IH^hOkyoB;>Qo9MU6sUBNv5kv(5XVHE+*;6$z&uzz|1{?l3XKmT?A{tA-p_WtJq%YQi41kqedKmi#E4j{eI zsT8q)eEcU4yT>4V2_a+UTaf^UmC61)q@;8Y&Ut@KXoGvTKdtb$u-5c4o44WFDHV{I zuCBLh-=lnB$MGa=c|exL^ujJ}u%%YM*zZXF4!vM{9jtn58G|PnTi^p|>CPSax6)4* z`QMI*e^)@*2ABS02Mj5Z{Hu{yBnb@!TOM+bGa)lbvnf1@2LZ{feAyU#Y~K7qjk9oF zpba+}AfLF%(9U#{T#tY%qJxM~VCN|#?0WR-8Yjo2|KB`-WXE#svtB&75|=4QCxdt= z4ZZrRO&}dfzYLX*TO+pe@$uzK3zC<9Q2QxIp$qpgYS78`%lfpqxVREDR+82?+*jO_ z%FpH}yG$0g@Qr~_7ov{i;ui)|mmvW}5jpcf7RxExmFtmoY|*l z;NKtJVqvYyX{o7`hCg58lSI{A7O@Yv6Yq13Qd!E$zsCvn&)_RL)?ehtjmf#q6LT(T zb@t2fX1g%9hrh_aLJCfbGaBQvbMlBG?|W%~bkAkSC4AOz(3ghBp9}UDu$+;*N+mHF zxA_%WKW?*K=+`9|z7f$9?ES|{C-A1k{z@T&knhU3s(t0}KZ+9HW zc0ZlGq6U~2a*BBTKYV(5MG9y!tfA40yS!59_9i(=_y>z{1j5j@_Y0wJas5~Xs-=#5 zCO*~^Y}tA!_~Md5%XfS^M>XacxeBYSVh_(L&r+91F4B+h+x|=3@4ubqGCP@Y$!q@c zi+hi%9zFV($Ndzus6>0a=G)Ku+m9Z*{O9op2XwO{ zCub^*BstX&(zmHE7_7Co)-5lR>s0ElQs``QE*t#-Tg@k77mz16|K!iPG$?X#m0`j6 zERLftuDzyr09jFrJoavrkS=u4Uzd zOMaFb(TlIVJaiWmaBQ}mOG$-;qwI_A!XXRXlzIXji$OymwrfTa#5U5*N3Db_^<(rA z0dcj397F3s*q*?3Ia{gtAN>5U-o7&^%Cw8p(J`Uor(#0X5kUk21tdsP83h}VoTCyX zO3qQofC zC!(;we$2WwKiz<2jfH}4UAhe$zT2H!b#KStzbgfB8~#s!V^~;f@q5271OEB*C~C>S z|MbQl{f&10FE9RZi8oZXn|}2wtMw4_YD9v`N&(iQev58rG1InHrRqk;#zW}qq{gA} zB!i~C8GQ08eY?+ZHyRrmT_Ce$`Ua)E#e3x>{%ha$hyMN8|HB7+P7Mr{c&i(mMCIY)YSpv?bZn_=JSr3z=R0XfVC!=hqC@>on*e zW}rq2d;EMZkaoTgk zeiPnMqmzt*;o&0_Ub<~QX&U$LpVBkwo&+J;9F^@t+b_2R={*xE=H7g@9H(X;mhdEgfUyudCCl?EI0{fq zd0rL@PTov&bw9c%Zwv69 zAsaRPLvL621OQIbg{{{kMWI>uh*RpZQ>rMOR8EL|>O@2B#fweg)jUBvff=g#g{U#G z_UpD<^}N)Df&K`tbuRQD=N1-r4{02EOaULE4uVH22uCo1LJQ^QN3SkDPd#n>`6f_U zF;Dca3wzGzK)u&nqdE3xU}NzjkUeu$FWdn2-0$S>deyvenK#t3dr!NzmG$W|GXxNB z5KnUW`Ta7#S^ZlJTcKloe9BPLe)#H=f{IE2Hja&byy7`pb1g=i)6Cnl;^CLJ(U|;o z2Hf8pzkC&{qtz;Gy`tyIA1rJLBa)-35wrfdjSpX_Z2Jp$8u7aO#XU5xU%%c>zfmtB zG?bm9w!3?ks0HMWsB!P#Uy!j?r*2+%1HHXQL)93aa+mp_1w=j2cN+~IsIvg-XCoNk~Q0gH?)1UX{QKiS$g3Vh?)|KVQ=c#*$};#2^ut zVffE~o**Vu#oBqvIW4Y7`AVJ3gc=D?W7MMJ}99i*$8DR0LYaCd2I4g zjpz;7dGKHc+NX!146K5OzWP$&(4n(93(7`DaR(q4S@!PTFu^<#N&N@I6qMKMS@VWY zLA~cFZX{G6s3j%d*FEzY@6WXY(Nn|fZ%RGLZyo?X%5LagB%gV;1iLv89s0nKkVG)m zI9OQDfg+{`b+lBg#0Y--0a=+^I?8Zcb^;a{wXIl2esc!5vAnp11PeS!d$x%`XcwyJ zAa@+=)P&TOT9VS*({A-JHnv|RN87Tcc5~|}1ahcOp`bbp+JGWZOI`8frMnSvvg6ax zbtuq6vuKe+Z$d0s|FJmBb^TwAOiV1$jX;fqeQRITKDmjGt{%-k414z`A#6&-z^ouW z=JORDH3RF^g&wn2M_(t09WN#-{)w`oyyTIaOx|RaEGL8aYz9RjNQ^1LYBUu&TceHr zB9Ph^a+6{t*3zO$i*SnkC8DMq7ZF5FLCk07b*@W9J|^bnAeb^ ze%u#gQxM+Ca_3U4$5%|jdO~qG<&ch0!9bx1`$$$cNPu+7Z-;{2-Yk zUK#T6N|DoOJva|LckGBoaGeYp8?6^^xk#3u;aA(M2FHZdtcp)5)67@93Ui5#Evg(HWn1Aj z=_o#SRqDe+*TUS~5Gn+YUpr;~!4WY9w}PD1>y85nTk$LdL57biBWjJiKNwQ22YA&x zBt1Ne_1uP7ueLrvyS%(?-kch0yc!xOo~pse^__$SOfvkt0_FEt$J4Z=xE&v`jgLp< zVk3y|d3%X>33L&dV?QPmBZnhXX!GeGZ3B5(*+#?|T#qOCt-EzF;DW+3)=g=yW3)htZ!uZj@{wmGl77$QFFJKA7{{3kfC{{R?F9*xbx-?s? z8~Mi!_JmQ)EyvE@9@DcNWc=SQ8s3$Q5=B^2o%OcBsA9YEl`G~bM5q}Wk&6dCKrN6j zGG{o}OH(5vm^`7QN1d=zx%U~Gz*M86qbBmuMVo@U!!X+I%scZln}kg!@{Sd}(mY@B zcs)H+9crRntgajl9phR^N)ISOl1nvSdM84K z_NnhMla5O7C&+CH!wIbB zIdHwBi)f+3vl4p)+9}t>a+UvWVZK+sbGdnLhlL)~bSt*ha7ahxOn=RprluU?pFQI^ zbxIR`Jf3%dsGhshb$nW>#l7+6-Lvo-g5qv zCAP_BjEJJ*>LRmfU7{v|ibq1cCfoZ9I-9=Sb*b86+~ywF78tP%)>f;I)mV(?cDD$M zAI(+0y{5X#urka&Wu6qo@%`nYj-EuiD3F&of@T-NQPnQNZ7&+!Mekar%+tjYjNkJ` z*$J*aNji|jX^@Y{U*2=^r>)OtVj6c(uovX~r9L(eLcjIofkxELsG=IATSEEH-3Dh|o z({)8EU33uW(iSCKyD~%C7J>w@=-6x812L0bu!7gFtxB=p_2SUV4>WUa@*Ks|aGl0& z)Ap!8?mRbBp>P`OpxupQI@OQ!=)Mk=|DX%+92;2Fj~E=@Aph1d z>dD*gh0oG*w3d!|#QJLGQvy;X#u>aa5dHxwx+W5ww#ZiiSMw}^^{OZ<`vCGVV6xu(Fh?iJAU_n|xe*8dK!C4nulz8g3 zcJ2n}SYRIsA;)zPg31?Vki`DYf(86J9xuBG5RPTdU)nlVJpn*ss_LXc5RXDip6*K=?=f@#!Wple|GH%YG{D_iXOTxIOX+p{CYXKQ9z zM6H$T!>ZcAkuM$W(Yu_v`9oy|d<1Na^VdiPfCXyabR#c+BB;n>+n6MMUd8)WtM<52&Ye}czWbW|Mx=lW z3J+m6aH|a`e{4dMPo3aqdi~~2F5o(#g&J23uJ*iqQEtHFUE0Rc7bg2`wP9nZ?2*b? zKP2GU4_E`E=N>UwcW=NaX`w#VguQnJJ||Ifl~yi#MH$9-8lxu3;8c>RpiyYOkM-wO z`T+nB5l67&PN6IBqqg;n{^s&A7IscfSp?S77cSJ@+O%6G)1XX3KblH?ite&w2s#H^ zo&D^jtnzot_f^1M!!1P$JAN3eeqZfhxU0NC_|1+VA@_F(&h}-NeMD#Clv%hKPj$4< zc{A_sUg?g4&^gJDMHgENAffRgtQ7FQytc#L9@T~!J|TSXjr|6*Iu7}}avU+SYfB5@ z(pH0-?L1P<(e51~lcBLMY9{Rm=6^bUNGOYx0VSO88OI>J|ZA1>Yy0ZFB zXb9UuXI7^hub$w_lks{4tE`S1K#W`HF_(qt&U$)E%f%%hL8EGho)`&rNDhR~rWhbm zT6)Slu9aQ5Du&IBtHjGYC%?d{9DP294zGDkUWsUrb}gNL`}Q1?8dW$DJC|?j>e!y% z)-cv73=TM2mpzNXiz=Y3uCel<%WJOAv^x8~|4V>%Yh)4>z25o!5!%A9KM)WvQ6C$)7$j zg%0xifpKHhdZ_oBN0c}`{$Z52$sO&R1;LScvRDO zy(6YXT}0zb>Bxk_=<`O*XP~lhX=a@xNKaE?U!M_) zZuFA>D)H_W86YXJXi$0MQ$1@ujcG~-8{1fK>C2p936e#K@qIOKblg4!Re>(ad zJUfFbbuN2F$3A_UdL#r0H94&UF2NC|{g1uC%aQi+Tkm|f0Yx!Z#zpXT;u zfwWHD`!;NlvhB?wI*uscG~h@VPh8dn__b8Kvgi*{!>1?_lL&_c0|VfA5TvO~4J_I?0uuwm1ujtt zc;g8~QQGV#`Q4^WQ_9(~GgmR*LjsKVD2c!So(3I}T}-TNBFz+C$V_93TYiQi%;3_= zSD=#fQdyS7g+N>tyTP?GF3l~lXEURqRRl(vrQ^_=!6v}Pt6~fW{K+!-|Fyj78VHY@ zO1!;k3=t$+|E>~6RCh4^7Q?ycF92O4xIaPP_lvqHkVo#feX}_JD517VbIdORo|gm- zfFV)Q*fb^J%$m@WqArgJw;qXhEX)*{@(Cp%9}czDL@ahe=Q<=`Xb;gtx?~D;i-b!c zt6Dk#K3;q!`U6r1GKC0ehk?5gtb`x)n#9n>hU}F}zh5{G5%R6ux6P4Gt!=-b(h|_e zi>Z5~56eZjHKf^vW+T{2h9Idauv<5XJE@Q@NL2ir4Ce>{@&&Kr2{mZ?gXK*lhVp$4 zukD^COeSUY*Aowf%wuf6wDb&tiqCcTK=HFId6Zxyh>w!dc!Ub~CD+K5%=PQnTfn5c zp$YS(!P^Z4VpmO8`3O{v0Ae6qX-B1URgw8y@RlPG&%Fn@5aJxWDP_UG&nEJ5o3M|j z0Uu4LhHX|s9aaUE6U?o83M;{Stq{*R&&sCkEBESFF4iKQ`%Dp4bCE`SBt6uV(NpRdCM;gbO%>1ixs0OGOa&6*PEf z3u^-t`y1J!fow&NBhQd4q#!jz3J%PI3hNR7^{eSG{OVQch?r-3jh(Y7nTrCLfJ-cP z!k#q&AgV45ohtSSfT~{Q;hzV+eqN68Hy0)&%o9*`NX8m^LT5eb$DUCxXEybAzkn9L zYEc;%#o7`fF*F7cvjyXqNGgxq+;RwI8=#Psi)=W2?Y*Zu%G64xT4UPZ2PV(S-st%{iI84|vnII&`&)*R# ziwrAgneES__R3W!^eOe0e0@cF4_m(@nleMCBjA1*0pnMA%Vdqr&->LGHo}-7upu(D z8U9OpN;~66VLUKiUOn$r_^Dpbi4aee0R467g#3P$zc2lbXpW@5Xu5tkyln!hXmu!` zlv20RKIf-M$~gG=vP=Iw^Ztu@2ecJGqLpJBs?shFug`a$dT)5^aB!eypHiu6iEK$e1uX<}72MF~46~js zQfQ(8h$zQ?P>vL)kao;E39;ToT81?ucrcRFfygJe+RXl4j0LHE%)zVu45|G8mf^oE zihBS1&;Lt)!2d=j_zo0{9Sj78GiR=$BAifFC2Q?-bLBV!xQU`URJ&g6M~_?S{_DTC z!Q%tgBrq~E?b|m?-4ZWdbS%s`U3g!`knbj91uT@A~u{Vl?5k9hDr`) z^+WQ^8NaCw(c6H|BV|VDe&(R()FT#l9v(#mD6@7(c+oXy1wtyhRtw|3m4yP-L=_KA~lpeSt-!;e;6PrFphe{H>+@D9mv}ewUL1fsIibT7J+>+3a%+hSqhB#Mld!+ zvtF=!da* z^e;@<^Vvy z6?>puG1&Rac?qI+-mcySFhmN)DbFjRZf)1!$^L_=H$X2ghOV!M-SKss&)=B zJm4}uv6&*L;mqAxpv@#IpiTiN)=G0n*Aq%OYGPHE=%&8H`akGmID7|ifTRMX|G}yGRU2bZA?W}1P`G~pgWk6 zz<>Z!9ZW>NhBM&?d6*o7GP>}EXn4!`{0@Xa_{^h7RY`D-OFIQZ2!)Ee<1-Wp7+RM* ztV@~*H-)@wqQY6!jhoRu17+$scq^DYIbhAC%%84PV8C}9>q__wXn2VR+7XX{P7`fo z3aHPqQb-;M%R%QEjqSCu_;*PJy%t?Ci~P-%<6Yz<0H+e(&#{bRG_CF0*(&Z`t3)BF zD;C#eIt-hK$_;rdOb3eQ)_-X^)TY%q$~tnOnB=o?!N{x6r+A{^oP<`*57pJG`0QNI z1*b}ZE66oS82kD2XG~g20Pvb3af>c!8vZ_)NsyOU31y_rxhVR*yyx&O)nHG6$*0xC z;(Z8YL(nS{=Ia#LsdX%)Ih7OvSv6-hUe2N0sX>0z6t4iIFDc*PAUYswB^Zpr7hK?w09@cEg~Ma+%`#hQYhWd+C6LCrz4nZA= zj7oB&_wkj{EykA*G{1U8^i_I2ACOq-+V%h z#-vv5BXmV=h}SoXjbX4I3k5*%%=?*JC~gO`>=Z73Q(ewuZ76WH{Uk*R6;6`x;0Z<$ zf#q_HNjDEZ@(#I)!6g@{-{_YvIl&_HTXkw+>e%xsQ3OS2egJ(Xk1>R(k-!2hL%>0u zOv4deX?f`ZW2kNcJN^CG3@Llb$_B6y7Q}(JmnR19R)8H87#3ze>j(l%9Wr!Q3W`3t zS~TE<6eXZpT?cB(?)ca*ug)kq;rZk9c_iu`zMo)n6%7oY8W;}XX^Jl50j}{+ zc-d7ycaMoaN_C`&I&}(GO2okmoAiD2FXrV@TbMMiwcSYBckaTM6RBQGCYi17rS; zTFZxap=g~0dq`2jy(8z2gcF3xCl?_UgASe~pcd3PBubjoVZU~qFT~Q(0^e>`@mGkc zDIrN=p)AafJLj*@V8I5ZKns8x2hM{TiqD{3WijZORE3Zn4@aP4U% zuIDW#71jZANF&6uS*Bf_8c%@0>~@R7{c8f9oQ_iv-V?zum*G@9DB>qbIW-wAMRCP< z+p5lMXM^~V2=x#6Sb`%XD8cP?Gx(m($Lz@Pcp literal 38731 zcmeFa2UL`2yEQzPXzU@zN{cvvf&-2{ny7smAzHV=vR5`w3 z?as9n3S|TBXN6M~%2!Sl%8IjJufpHxFRwPhw^iqqjw|3hzGjYOdE(oeOFy5nqfj`n zlmD0bb&r}*DBn|P3O{N%g$}eh-`i3fCf_dHa!`7u0^4^#t`SiC@zS3^dhGF4(feWN z=F2q-M>KA&wAWiD_Pxyw!JWJp^d5D5t9SUfPGbd|qZ&UP5IEMnWsAh5W=P~i#?cm#epC505wZ#<((SGI}-IVuu!+W9@cQY4nw>`wI5gZri~6>B0Gx*{mOS@V0+N zet5TMnK1c&_}Ch!W#k7D#aHXd_dnE@lc%G2-THQ9Ir+i(uGcp5{rkuNiH~RHX;eTN%Mi^EHa!%sDvdfo^5t4Qr#n9p5WoMX0xN#tDU&U(1oqPTJno<9Vd8*XqPv4Yy8bxR=W^?K38^MOfz=4!89VFs!eOZ&P(rM&!xH1>WIaIizCpRweuljUHkV{N}A8K=}%>)^gE;mixZ zjWQ#z?ulFKlru_#gp+D3N}oQE`!3x^tJTqj3~!)Y5sCI{M?1d@Yi4QcIa< z=J#2u^j_I!=yt$Ysz3U4vN5fkVV$?&ki9V3WHXpjf2<;0)~MuVVT`}W;3 z@|ENZcj=D_tmXaPbzWw75f~S(MXEcaUI$#Su}{@>_Q0lKkQBlMN#E6~UtX zT%PlzPV=LMyLlctkN4OtFzXB3|NL#YynJ4>i$67{YWTsUN007_S!k8s7Zc=D3Ekcs zYNrzFHdbU)$mytn#THu|RQ%nxZH3dLot>d}4#q`6JxEC2N1G)ip_R zm$i+J%`FZ|k;KgY*wA{%Cd;%#)9t@KcAXj`EB>Cmr$_cgmC^yTswby2Y)#ze2GhtR zq+`O3`fFk;lTAu)e^58U%A4!x^pp49X5jSj^*0$(TZIicu5)`h%@108Dj8bfdK{S5 z=Y?FZ3XHl0qd$NBiY)VchiuL%C}fWous`n&cP)rmoQT>k<8fGp1)^MGpUBANFvu7I%F$Y0s3T#f8#EwZS(k~+N zTpxdhVW1zoy)$wlomr>r-_g+#DuSI{l;pX|f+SscVo2XxX^;MrZ#@O|g zhK9!ZnWzX4sr~hjkNb&~V43q<)Wy>Rg$={AF{4F$qeAJ`Pn)dsa$b9A)c*R_Iy!DR zis{f_vo~18L@{H|Y-X%mQa-}_XD+vqKi9T(mP-$p?)4n`b2Yzk_Q+4kPNN;hhypY^ zwFnD{es$e8z6#kXJ=^y`uI|1oKVB-TR6Fx-x8DWTamuTo%*qn&thPgPLat`9LMG=Gzfe%CWE*~n48 zLR-(~Xyf%ie_I`-Yehx@_Y)fEzO*o<6cCKnmzj&;TjaAvusA+0J^hem*B?i!l7;i; zZLeCAg;|}9O;tO)y#1zd?$5ZWzm>ep#3zxGfb~wUlLKeEXN7a8J1%D-xaJl&1w=(h z?`d$KYN?i*V=}adJeuG02IrnBJXx92>bRImp1pETBKH=e~vQPpummhR@F( zrQ;=uo(M&88pAFY^@+-jsg@yDOSoAL3s;KQ{yVg~?~iCD>IDl)ST?@o(oG%}kIJPO zpP^H2I&Vunt?-;{Y75jA7Zw&q>bWoHIh4_>7?8n8%&O{E63U?7KWKL9y#9ur2j4lj zA+*OgPmvL*ey+cpC){JU?^At}VMe#aOYFKr!=>5!iJ6g)Z!GMyr|c?ZCG;1^N`%Rd zV%KbA)*D8s+^c(TR7gTyaf=^&W*1$lYSi)O&6^|??tRfLJ)FJy>eZ`kY$QkvY>K?} z=U3ZEEUfBXn#)@Xd6A2NXEWOu6;&~uJzg$Qo7H-2hv>PFugN=7Eb6EK+~jdGVQ?8G zRRIXWYyStVl5<_}?$``GFFb}g^$6Q+OQ3b-qobMGm;0(tF_tExL&j$?vA|?Xmq$Y1WsATV;9blqnJ>#dpo?)xt>*?49Ya zjkD0x#un6b8-3e1Zqr$yDcOAc^tRBT`5h!#_f$j}#x@5TdISq`yL@^W$#)_ydXbC| zxga90@VKegSg$=&gofaWJZ#hIZlBzX6iU0{%52>~NJ2p7c%=6D*l3lKaDa{{hUE#P zK2e{Yjg8xVLXBLN6~4xM${ySGmXk~{T`0e(WY!g+gBj+TV?h=Jv8zD2uT(7ng- zHoa!slKlabc_O@16&-JHR@aA!Sp?>KbO&e$W>_Me$jo&4kCqrNia0cw1YN+;TzpX` zZZlDNybxd{&|sH(c(tm$th*mV*y)!S)IyPzpAV)s6^J)o+?+T4R`}dlSCMA8SngBz zh7TS!M-5}+#Ai=dJvw@`qt$izwz`TWugA!5+zMonkZURvbkn9ADm=D!7Tgw;*wgU7 zIm52=zD42}%9)006aT?!}hRM2h=$H%j@&#feXO9#tGl(~Q$P zs*qnAEZRBnqQaNcvH8G#aZ46FU7 z_T8?&@nBKYKmk$5p=PI^VAHDJdObVkNXvZ{CPDgJk8`_U2L4*{91*)dJ$%D{!(17C zD*&gL)@`{3_Ohe-oBP_%A{yRYvGLG@M*E4O7HNbvw!B4VQYTj7&7QK5Ac5*s%cjfD z3r{X++E>x|-;zm)SeiEtb!>HxtQ$^kvLY$c`?O;yLf}o!SjX<-0Ljg{oOY!LYR!QX zx5gQ)_I^-cF0#tnWOF@vSKll~yZaWuc3mf3Nr*-_ZrN_dAA0Ght0oan9XEESSiBLx z%_W=pdddv>e5`=oS|d|fATA2Kp{O^)Go0k;J`2Q_$o`D6VnM-p(}hlwW%&hT66teM z*@bv?E%PkIm+o69KRDe=dD{PD*zetsZvHg zfgIB5*mEX!b*nE@1BNCJt9esRs1VFARG|SlP9? zPX*{KMaeJh)o;sjozm$$?rbi#YS_MBKgder;)iFq&EHfW=MJm0uROZVQbT{s8Ep3Q z-t(wbQgl3v)-?M{_S_S&9RFbC7uiH$Y)T^+#jEVh8!wSi0OiqM`6Wpn9$)#z=|f0< z=W1g$4_Gy)-;;B9QHal4dww`)%9q_b`#`wo;>?AnR7)}o&yiRPP|t9X?ct)?apySK zj`7|KRqHqLHwf0wv9x~WwajMv<8@Xf0+6%8#}qWCSQr)2c`LIVhwq7+s>C$$)@OZs zbhD$;B1t)7_(g@>=y=4Ej82*rXHxcpSY}Av^Vr14QHlweQRtNLJc+`q}p+9Fp2aK|#-@FLHFETD5Sz zLY^BasXMik0AK=h7w3k#e_OS|$D}0i?T6<^arV5L_H(DUEkq~eYG_(c?xN8TVd6WB z_#;jj&K0S8e7i6=>oRnyS;4k`E6jm`XZO#xo^ zGZ_sUy~(w4+Jyk++kqTaJgoqnMoUF2NZPY44$K~0B9{#4Z?no8tm>|iU&^kT1zg{! zFEnG9q)MabBLR1!2)z}Tlq=ODOeXxQ(HiZ@R{n4&9{1KX>$clg7S7n8-(ms$JRB-H zl-}9lz1!FO?3-&xBjntRWv*)!B6hJ8V6UJcGyCcBXqiM?*ZN_#Kzc+Ay>sUZ%E=qE zCxT)!GIM{G;Gu^(wn`ENs1l51)P?9TWS{b$AnH@vT#2Wj^Ypdmxa#!ydCnW9G-+yu z$Sq7%pRy-aky%@=yG~-fppHM4X5jSk>dO-z9v)+;p8^D4U=p8rE-gmRo*s)$r;942 zU2I6M@z5D2@gT+s(AXp}E|#txXX#X1AL>iQ{mqq6zwsOWL*9<`Nz}_}+Q4!Bi<= z?$cvGIziFhVvu_Rr%-$p%PU2A%$k+Y8Q8qtV$!-xZLI&#Uw{1-%aFG( zQ`z||YfDe7{;}-rhd}XxQrCER`mh2-fQv=aW+IkmB3?FswN3A2xGG)Pw!MGGmO!ew3g^5(mROBSxD#|o@F3h-+u$KHLl zpJBQEt{cFPBtvP{8q22CgwY>1*CN|YJr(o}GZ(vq3=^N_<}D0bn>C=U z?}``1#ibmFLK7p9e=?ZySbn`*M;g~US-9-0@3GO-Pf1ydGkUY2Rr$wP5uS_ADhJdGiQ!WdYaq7?rt*%Fy2;+BLKP+KE(pEF>lg9(!7}qB zZ%jPS+9RVxb|+?!6$|84V9SMCiC$`d6>5)KvD|qWxS~$6t@y>+SGu*w!AsBqN;ITO zRZMahXBnX`+l2H^E6EBPqOP~aC-%f2_fUk?{D;i}QWspz2;WH(4mPhM#Fr5D(aK=+ZA&{Zl3e z8btLMu{0GEJaLf@e^t2~-u7=O91b7OcGqd1Jo$}dv(51ghEln&EKvMf{udeDfw~>H zC0e~vn-px69yn>0(fEU1Pw72bR|kc7_O>H=2F|0pr%bPyh z)!SRIazs!qXuR!sD7Q(;q(NfKB<|>Deb_bpxnt5^~5KY z$X$#bKY~(21{nfFAe8as_1Ny?zPD7uq(~8vrFLXxt%g1}mrLHZEnD8cyR*A9cW&tB z0i%NaA6IjJ$Hw66lA8QrWR$lsDJZtLdA4Z7?nhN0ZR-RZuG8}jOI%u?(EJq{S-9y( zh>n%d&c1mEwqkyM{sPcru%C~Ir*>g+fM92Or|(UD$Ci}RvwX4U3ti?6OP|izAAehA zGgic32yX5i$x|pqM5e}(SGN)#48)-jAyc%>&ZQziMisWUEiF#nlXkQU?#A}m5&kgs z`l{uW56x?w_Euk6&2{?YEM4?Q(0U_r4coe&4NJ7R#NED!9=&F=`!clj^g_qyGkU|7 zf&|w!@$k$}^r;WYYw<3viI8!&MavKE9vc&o&nF?|@(fb2 za9T6M)(e~4an~{jj^{t^!)AMk;08W)Gj{szP4e^4stzbAtrBGS5HfV`4!AsnK%MY1 z@yI6GmsP3Sg#{cXDb5_)$1R7J?hEW_4eSLRa?YGb&JQwI*Jqwp594n2m>ax2^X$#W zl9}Y^`$%GPL+Ctk_Vw?i_Ry*?6O2rhm$1)in*zj22MyWz#9zhdMY)t9=)VjnRG~(s z7W4M@obTk0^zEg)?-3A)t(v>LUW{G5G>q%qn_d)Sv8UV_+;88$ja=l4js<6^kUV6S zc8U~A7=T*{8-=LjLVS12gO2$?I0nhDV=dm;SjS~G#~Wt4!#Togakxc|;S zS99M2VDFXmtT?JL4`_AEAjjn`qMG*EFlT3HNkfCc`Cp$tNo#cxTsYOqFF!+jY1QJg z+jRxynF-s-si(z!$M){H_tLyJw%Jy5O-gJmfcTcgwvin1Y=J1_=oF7fNj?WJe)s{E zgIZ4}N`}It+w||s0Vno$f>pXj{>H#{a7(?Rr#rd)K6_pjeOE-%>Tb`uG|#vyao#Bg z3hFaM^8P7eYMPogCh|Pu%7*=!o7e?}gnYkRw>2N(;X8@uOGmD59?wSj>9f`Jy$TSe zF{79r`{DC$Yt#tCZqw{LUQ{vJRpcM1xl}MgrcJSRO)_d>Yx-2Y68l_6!`N&Gu7tX16`magopASI@C_UN%-2H{lqpmc0o7;=d(77dQh%h?%b)p!M?%ITe&(YbKnT7_Y)!F`%&y~ zB|rqc=i}0|wBqr%+YFzSG+lh59FW@J+#A*j+!QnBs$p#8+ubPVYjMtK%2Qdh+b5OY zI|M9z6nP=I_fzB0o2OFpcR5mjAyJ4rd`7pw!|ui2iV|M?zA9c1WU2y?HJog0$liWf z%@RC3rqcmqR)uXRi{kW4r1**0U=}Td2(CCRvQz5vh4q|LFH7yQ1PM#69iv}cnQ711 zE@7p|ap~1EHnvghJyN2ci%Uq2)siJOI3Xq@n8-3m!5f+96V+sPYM5L7>4_T^a&!CSfFwVmbjkEUq0#qD0eSs4B452b z1UPZg#P-M&A?ES5{m}z;b;Wl%cpixtG?OA-&tsSl-G`HE*69 z%CZiK&$%)NySUEL~6nwb1VAxb>V_#FZm=qK17Od4E7Zb4zUWU^1M&5%D=k8_)q{OB~_bI}A3yP4ZyI zIf4H)LKQSRKVDG?22!cznjA{ebQJq>wLSeho~{=ZYLh~BNPRm#o5>UZr0gNLedTrJ zxVHenT!Pw(C=J1rzukn4oDNo^6B)=`qSa}4?o3a}G%!A?HTO@sOF0aj9!#|?eez>U zOrmdRY;*z%Av^q;0OPbE!~tD_Ju$;_7vt&L&;pqCDW^z(E*4w>T(oDao z;(Sp=rC?I_sYtKvu0J+q4P{!yFi{ho2#NryMYOAwrmVf+efM40EzK+xSvza(Ua#SQ zQUd-mVKitx32L|4C4?5o<(FR6y#JcpZRAISZk}Vj>o5HD0ysO=Auv1Z(Z1E3mruG( z6#(1ZvunP+Q%pc#NHTN5sx`Uale9ZOzc`hr8X*sP zM(y{ku=B#SBJpJ@ zD1lHRZb%T4jNEnZA5Mb{w#%HK%3I1&Y?|n=6$CDOHa!g8b)uh1|8r`#Jw*mgUfhpL z^+m;UOCxVKb;iFa_gR`+S~Mh;H&he=KUxq-w5iJ(sA&5n&x2$NP!SKL(#n7Lk_TC= z7%+p|ju3sw(A3t~j1;#rM}92Yp_*>|wE-upXyrHWNghGGj?*tB($o@I)vDgS+1e!A zx^3TmmyeV|LfQ?W+?r1>v~5F*`DFI(+4D2Mk6!A;LW z{!n^Vvb7V4Kq+WAlaZ)cQV}XC+UqbvCz{8h?z`&ez&9q1q!yyE>T32$hu|l*usYZH zfgON=mRZlpYA&bm^@W@RG~63m!@tK_cZ^WH7(lnHX*=`4+jS=jN_uqT&d!)MDXLu5 zYBL`{l<3>oydiXz}i@@VCP?6fZh6wc_7>sJyS8l zHwlGI9uz#0a^AtclHsFZAq3-%dP;-EtgN*roe}(--NwsCLG>46rJask^OKT_50iUo zlv|RaXwWcIpSPIu+NBi z7Q6z(Ex$A_-)F(gpY#>wfy>J5iw7LYI{5HA+0O=89=0WXv~NkVoWTI=v5=0PLuQi3 z9~={%$Rtc&)REPY4Y>Nc4j6jOm=Lw^^_ZdOf_?cjHo_8wr6fp+U-zhS;8ygwbF!i9 z6f`v(C?`GHS|a%^pr1fwaP`eu9pXAigC+wq@zLzb5E5#85Ng3wWWJ8uB@an89nod? zxt2qZk8{{4l+Vr%9kEZmzh4!|!U#N=nd>6foYv{wu7oz;Id$7;Kl-xjYs8iS%pb3& zySv{rLR%9#s*1KQXuq!3N6B409XI&He_l}s$bJav$F43YyKE5xpTJaa7?sB2QpY#& z@855RU{1v5Cly3+8XPr4O`>DYT^|@AsNFM>1(mfn{Fj6$APk0Wt{6~MQc^M#CJGBhI#j9H?}n04fv(!faW8NXC4J1OpifA9 z3JT4D!_N`2(*ZUte)GT|u1ia=fMqLD$O zT40H1m)c@DZ|^#C?McP;^*c^Q`Csqq?tc5*Cb@fZ+|Xy6v7}5kt|Pi=&-&dzpf)`? z5oaB8MBY4W6TrS|rrX;s+IQI}+!2{8lMM4tMGS6OG81-=>=pJHy?I%Df)1r@2T?1C zalj;3+daChGhNEub*VldiTtr@gq(@5>>M^jcd5+Mb@};Sru*J|&ki2}RaMnZ#6+j6 zB3-KtUVS-eROg!0ZA`edp@IJ*CG!*J)ULegI+x3Z%YzQ6M20{1R5)imJRIUx1cue1uMkE(}ArB2E`?xRBX1_UG9VYnumH;w#Eu3 zi%EiUg1a#vbi3|BYme~|0+q;AjE)v^_ug8U=fdnK)pRpr!1Fg7rSoyYDtlbbGhogh z1bY;-1-r0s)E)rWjh}z|E6TfDzn=j2tNtw;gz@U9%0T?k@6TqI;qR0ZSU$BeS6cIp>kK*>SsA5IAq0EKc&*DlG*M6_dYEZK{9hu$@p!f)Se8$Df(zBp>=5Og88HP^U+P5*UFZ+LeA~#p<4k(R)U|H292kd)=dp4y3>Z5ZOem zPtMqsbchQF_kwFG(U?4=NT=?AUq&SY%1o~@`0}DRo8&{AFIOwmAlKT~<<0iTMfMXj ziApf{!l&a!V7dKaMv<2sd|6H02)t}}2x95;FvHvi5AGe7Ik_z&BvVFs(+&};X#~)g zT+zJ)Ms!fZhcZyS`4bk|bV0H=RNYD+(?vtDs!CSAe~>1WJ@VRketNVhTJ162ms7eY zJ_HLbpz$;a0;t1$QJiddz^w90BG&qY;Ne4Af62Bt%#n{pmzySbJ%+GV1h}CdyGYy% zQO&dCZP)yg`}LN#i(fo_+jA9*2BQPGbl0>Y9CiU>suE??7&1u_k+lik5ZP$cVqdSR zvqbzB@U>`vi16V8*I=A7tA6_QJp?H{GnWzD9U>@+wH&2Ts5GATvOUYJ8wA!}g`OpU zo!AYoEGU8qnBlr_?ny#oCqWRA*3ycFP}*aS>uA z8QV*Q5k2gcmO85zP6MTiUIsFJMRFWgv5~|NZ6MXEo9ni0cD8k#DBb&*QFwhJ1U% zpnEvg-qk<|i^&S$dTqrT_+l<@$5M}ELT=11ZGYbN82F1t2@a-ufD|agV%6*OXc;*G z8ftBu^&77=DL=A~$l@>Gwl(D}u)n1IR56#V>*R@6r_S2~^@bW6cbhN0*$}X*o|hX; zis2Z7QGK-hZ>u+z$HXC!i^Z;{czGUBSfLxbYQxTaZ?@eXNX$OO%|^743s~0&ty^V0 zCLVLY&A+j^6HX}~QsJ3qG1F*)@x2fUZxcBOf4`I1;IR5g5<6&7w{LcGkR6f1kJz0I zFyprTzA=LGi=%!%=5AKQnGL@Zq@)HsUdq7`tKCK*-%DKU?W{4%PN3R$-yCM9X*kbyHcswiXBctx26T z7{@{)zSikl-}YFTJqk{i$UBvgGN*|k21LVjJooPAhFT&}+CN7jiOetx3MPM@aP~LQ zPtL(=#fKbl4zbMFA~AyxxNmdbT+6Tk$Ev5fbC<9_Snv>X%DdopGeDEt#5l+pq<)=d z7%ttn(*qMKM1(|QsUoXi@(uNF4GfV^FsXF}n}**2ZFm$`N7(4(GaUwZA|6>0!;lSR zHp>aiPG(L;++)1vz~xWxiSkbbtpkuxKY$4)BazzXF?WbaI*`K%^K>!ega&^4@!Yn#^SXN>;(wW>~KQ*Ap!0> z=F73;IH$eFWq6s0iw=93m}L(2+%xJR_6$U~9}K`%k_3Ev)~@c1X}64_{6*A!V)ftC zig&zQmwE}1+J{g1zCwQHFDCHjFq?TEWF~uXKpdE{d9y)6K_MS={s89=jbZ8?uODRB!P9?!19odnG zWw!iv*R<~?4m`+VA)}Gxqros)4!{MZrl-fj#x~fL=BKVID@Gh1tQ!gjMwA#9oWPW?474=>n| z`+X&q_?M&EKc3}e{y#D-|EE3AV)4VPPBy}3*PpfK@CL}Yhsap7u?cF$djpx*+ufwR zV}&0e4+RAhq6z)$V#P)|Q-!4Pu0+>yLmQS6BSY-4vSkjlPOMSo)w+{PV`F24Bjn*B zrF=!gvhOYGmNqW)>Mw>)aBV~VQ5kr}ys5PQjH9of8haBz`u|I-_J8`x|IZEM(bt$Q zUX7p#2%U*N@*TIw^jVSwT&9LNiE3?`+VHg0QY-#6+h%^1TTn~$v8NxwaL95jKudmv zK$Vip^0XheH2;H2JJohOdOzebUs5-LnPeyYB3VgTnw;P$Z+Lsd4@7uI+tf24!PWx- z1@70~3c6YU^1z1!1T&saF=vr^<#^wzE0)TsF(l5_IO#rEJZRZ?kSP6ZY;A7i%JF&` z$>F*y4{MR4NK4mp3mY2>Qxc9o&nw{Wg%6in#N5_x6S`Hlr~k_SVx$KB$AjW=f=)#K za{~L&Nu+UvsN>M4VSxoHM-1}B_f6`VJ_d?$fUR>_HV0rKAqGH&_;o(;T@>GgjdK9c z*vAlSVNLu4p^M^1BSJ7n5Vf0>3G@Bhd5TAxy}e0VBZ?GgDIbu=Qtk_&6mKJ+i(;(9 zHop@$uh~P~@!sB~I=9Q6i7gT*Yl&ii_C?08ju3o4bFU)Tj~|+9wo}TPEEY^c?DAT+ zLNUPt_A$7w_@AmqaDCx@>z&xiyS*}>HR$~*jh*Z5{Q=5yY<3xONwcvb84K7gaJde! zv9S>v?wBR^E`@S(n|O$zlRT*23n&E+npIJuANfzir}iyjL&@-OJk;*B9pdB53za;k zA;5f89v+kzXU>Nao0NAmt-?I{^McU-uy1)Q(qRE9Z;@&IQ04Ivy@|Y#lvl(-O%@g^ zT^X45IEgBq*#g34H&0bCpF|CIskb+BVC6B(7Y|>*e(eN6P^|9>2kcvLC%Qu}?O8Eh z&JVGm?gXVZG*g!y<`k67KaCVME5S?AI9jwM3q3-|Z~roLb7p_7SgXT}ubt{aC$O;* z!jAxE57&C}7*9!sSNu)FV_GNOx!0l=6)9OH2b($@f#mQ)T zNQa7AtWW4CZHwqN?O9Q{iAt=KS*ul)9d{!lgxkM*P%yvo%0=iiiKD3@M_*+bz=8^U zSmZ~m{LJqsNlyVGn^56(0+8MUeEc)V{c^n-bT`D*TQ`~MILiPK2M^}oQ zn86OJ2ghM|qBXKGWSQ({<|{{aVz`8lla1}UZFlk7q^{LobQ1Rf_fTa2dIg+Q4dN%I zzb){tUqar6mwe(ky(EA|h;xu^#0;Y{s9yp+Jb^;`_j8Hs^>ntA4XpKnJL<}ToOyWA z1@cU->0&glLcok|y)bqO38`cf_u`{AkcsI$wCX{z;;L*KeVTJQ<(rSzj|3Voo~JZ| z*WBG4H`rxSbtXDgI)0sm6kaK*4($R zI_fw|-$zP8LVXPIiAI-)nk5OpMgqa?Eb-q0L}Q;qF#&M@^Y0JcN#*r_e@i$0(pf@C zJ%_WKV4wjYol*D!-olXq)VkFMm2|-r8gD`^x9&Ur4dLrBf<&VKhFmOrveAOj%>>)8 z=a$RXIIZh%8B6TOVy#=F<>8i7E>xZG39&v0=9gfZWF8D_*poC^txa3MelSEVq|BC; z6oDV$)R?t;&6k2$SAQX^OrZU?_vsuL`xD_ZlEmapxC?;KU5g8COR~O@-fj7RnS&p! zQ`mDx2?gl7=Wxsr$i?A4QEBsP+3LIOHFd@zvEPpTB#vYhy?2>#&8B6-uN*_X(#*xP zyd3SmZ@+pG?Mo-smm5TLFE{WS7GGvvOJXIfqv@jAf1;D=|6+Rl?`_HY*Mt0*cDVhI z8%_e=D2gQYcXb!F?y}e^K3~5#1@ZMxM1-LqfU45YgBSh~>h0}44x)|mT#Q0#+n`(3 zra&}?(bVN=3H@bZ*Ep%E{r>CwVte->P?7ewE$! zBhmhc`6y!VQ7la~y+tx6T}ZNNUwg%)4eKW94B)9-u*v&^y$&zvA_2Hdob@*caJrpp z=e&c!i~pwX#{Y41|F6hI`M=aS`~QihwSkef%zp6dHUsa<$Hxm{ax(%L7kE($uKgVN zmjf{N5gDCOsT+aai4hN6%!9gAR{b5V9Af_}3tm*BLp@GUPj40>{0c;+e4w9mu$|l_ z1sbWmoInpe2AxH;y}-2>ExrJG%chGfTG5vo8EX^Q^+MI~29k)r!@cQUY}6-xw}0%r%Z589!fAY+S!5ZC~}| zMTsbtn-06PbTaR-3s^l39g?9&K73^^p|#Js6e-y1M%P`}b#E)>GD!g+!!bciB~x!x0SCw#^BKdAWtI7CqkAK(m!NEaVxU0`O z(;;GIZJjQVuo|C*5|lZ#;_l<*3kyR&QQ}rz zL3uwltIo|(J`#Q7RTgXmL~NCWP?ii5dhmUCW@aWpvdn%czZMosM~=MFvIv*wcLeP% z+gcq)+TYsddnD4Fb28d3uAo%@d@ewSB&dtlhn^eTYac&<{+x08T3ub;#pd+qe*DWH ze8*%m_1olA)6z`Qm+3CUMmgM8UFdK^d1Wg%yb|PLzyfCoXHL#nZzANMx}jl2jCxd0 z%dm$d-56wonVA{Q!oni$`6S8-6*}tAr;L>FfxnjB%2upXVbYS6R!`bUZ<05dRUd(d9N-pxQ7c ztBUxzoU21xCmvZ4vn@z%|C#aLUdSSF?t-D|0pnYOiGDy(#sA|EX8)dqGo5t_T zdNhaBOl;EfT&`cYSy)*OG1$DgE?m5L0!x){485=+-NuMpb>-o`{R{??=czPZU0vXp z>cY)GF8zR_K9_5m*C1(DH}o6v*@+Hizey%T^z&g0ihQ3dk?#~gbC-MWem;ERklVhKDOZCxoBYd_KP6>bIXiU@h;zKHxtbgJ`*R*#MozXvkT55*Znpbc{8j zdvqeNIAZ34VTsP2DYJISsYCQc|ByUc6uYi-(BaoG2EJ{0X{H|VnDw5T(BkF8#Jg~7 z`vD#z2BPIDGqFAiYrGjUn6Qp3is&AMwiBkNrkb5!-lMpD7*wDw!r0SJa_AP1qJkI4zC1x*w zY94Mm*{UUzT13`w`<2x_zpfK{2Xk-{wFu&r#@j=z4-ku?zjB4cf=VMM)a=QWq(K*V zv{~rO%b|%Z(?bTvZE6uwSN{11_;+||qy=DG&f(WwF`DoYR9Qo&mqhhDh$hjDob0)#;B{PWK zYwTtn@dnZ8Aex^2^WhV$-7V-{Hx1{{ArMzJGz3?8E{@aaNc|@8LNIii8XNC>F7|q= z(1}w+^Va9Q<}a`&o1DSnPhGcl-#eHdBVbxNMB)*>3#Wn@_xJbHscFr&N^lVe5vhB@DL5;Ouggt5OtdE*oH{x>bSi1V7Ihf-AW-l*hSF9T zy3i`JA-_~WnFwsM%GwXBW9i+!>b4-#&ckA)Z$Su+gW}>n;7?0&(1cFyo-gy1{3?VB zlWoOuxv@Kz26^yOX8BsMe%Ia|)8&1QlXr+kS4`+dtYSq0ic6yQ@q_q!l8|6&T1Kb# z4Gc8LOtA(aS~L7hDA7m?Z_feps;;c8983YF)(!4bi9V3Gv^ewry4+p(>kkrJ1qNa~ zt8GCD`O6YI68ddmkQW7W+xmG(-zU}BYhB`dP0UEB@=SoZNOR}-9X~hY5bo$n%#)yL zn7tNen@&2x@R@B5o!>=eWMm{Ar?@--W>HpFrqP*)in^8cejb zwr0&8wFt!8KVRnxNu8fYT;AP~&`J;rGWFA7)+g(gW%i&BdrW2hed!JH9Zy$tTspki z7qPh4aJDK)1ZllAOj?5I!vtTG*hAtOh-y3N#14Zq#oD{6ccdU%(Lr@3%Nkfww1J`ObqDh9OW=q7oAmeT3JLcfTLj z_|^&T#3fOhEy(*Gv|X6(+_^J$T7>+TvY^9jIwGW({sV0+32ogSLgX(GA4Ti{(4&P* zhqJb|mpMKp>O?S|YG!4X*zp4~`3He3(<=d6=cHP@2fDV12$u5}n_vK0#OettIQ8ovt7*!^9bv zzMu7b;SnG1%Jy6OBBKBPFJ!Ll}x?J8?JIfFnF za#5Ct2bzSt;ruT{5p;*R>xd638E$K85o~=QNwd<-tJQKRYrkZzDEg}`>~z9c)94V1 zyQSM^RG`u1)MgULWZ1BqDKQ1cfY-$RFrku5EJ$ z|IBnqPJeV|bnbA-hx;L%dIluZzx}$reQ;o4pfVTHAWVi93?QhktzBPlL=moD=h;$M zUvG*8;C^TWg*PV2bJT8I>k$fNwth+>r|(g~%2pDGF}=kky=9}l-?NBG&RL*StzD`^ zVz);fLZPg?YGZuY($4lRBB1zDMmtQiW_;VM zvzG3|vf-ZlP-u(82hNYQ|JkIMg(N^++cbJw&Y-2{E%i{GPgS=$`{aXb*RGM7pwWTJ z&x5u5J4gfO+^3pZTdM+wAV{P5fpM*=es}h{#-=9M>@;8uV#h_ndx^Bk&kwdOT3R7J z6OBPPK^KB5FhU@793*~JVYxmwKGi5~>UUFQM_%ViRkgMXjVyom#=uG8h^cAR)Hf8b z{3!+LX!*l0?KS|#DXf($*FAvI{ zh~&Tt31$~A{5;ky89;oN<8OsM4v>V76czjTLZ4hYJdPeeKRw$bUYN&}=rWE^?``^q zwGTe!C`PmomXF&wVw)#T1hFKd=)D1}TJ0Hi-SCXL$;~xXSlJJ{54#$Y=1e`u7JF2T zaklHp+C&yR)9#KoZc#`kdN8bHboxrCX6AzBMA!G`4D{sN0&SV1`{eJZK=5K0zxa;r zIL94Kt2oUTogx=gdS)3(6|5Av?KeLh*@cK4`S&P&hM1+$i$$@nAnJL(LxxDxvFo~X zya#<4Wh6fVXLMoFmt(8M{v%7G!MBH6gqOb{&PvBY0=3X=AwnC&63+SOC16aqr+59n zS>?k+CIFu;owO(a8{nGzBZUQE9@@h)rb2(2lA&e-V7m5`sLyb zzsDvtDEy7OuzSc#rqTV}2aHhZwG1^PV!C0gE7P+(IftYFfeGAurQ3pMXJ-#eNc7^k zhf-@C$YWJ63-2~bFysUXW9(Ar3fETkzf!%~FIyy9(E3D<@Cdc9J97qVGdVSloR$TP zS260wU~*mtNiL+1%yAH%xhltQZCCs26M(@(h_AK;E%kxK7>IVIi}yo9>;TL!rdZ)P zIZ}kEpkp!i*cT)j@Kq~S5)BGfh|_OxZeta|;MnjcQg{>pk|b7u9m1*!HnpZN>uo3f z1vX^UL_nF4K#snsm{}#(FD3*7+}@$(0AICnG9y-vFZU8}@%WWZZr_tfU>HngCRz}V z#A9(hLi_K?KjD21)`u>%v0`#nKT-SPz(=#_%shlGAIGNum56AX( zBL(*RWLdj)vZv{c<7Auzv@ydarlBMp;cQDMaX|^~H!vwB8;#quIn@P%c zGXDNlVj&)bt1a^%%gC`jR1zDc1QtO%95iF0(bsl=h+GP}n6Ij5*71ojDKXM%6=1|#bTv;DS6GiVZ_XqeUj?3y*u3w^FAnFOYWNGmD19kd zYEWjZp1DV28p5+2jkms`8Kx{@l~?c+Mbci^VS z;|v$fNaZ%j_ zXIQmG!DuI@qX%R9pqN-Us?`s<>Bk07d*XtnREX za!!l;LS0%GrYOr}9-6*&9K+h_4ZDc*SQoX3G~NAO)lCSDAn)YR=#DuI{T*oPNVBj) zk3O~N>~+pS@B%qaVo~8P=JYNUTx!;z3*ArJE?b-kvn@>US7TAC3RI&>uXUA2i`HEm*dRaC%KjfIFM2FomSE7?OixYx5YqKjn4>e+|fUpMb zOf5of*F&I8Q+5T0iB2Uh&VSxKtQphPA{bvnaW-h)r$%hbtk;99~}a0aoK z-i2&1Q1%&-E+{4~AC_(%^`#b(3K(7tQ-lN>9bk`ODSZZ6^;T1k4h}syUafC-rfs36 zjd}iaG3pVieZ>(|7mTxd=w0k>)5lnpH3qx^?fSOpj^Sqo2vm0G;&Q+@RKS*FhtS4L zD@WTB;a~qAm>nOQ*vau>#Jx6zdU789W##+Rnyc&wkz3zkGd3hl5}v0F?aWyo5wcib;CSn5b0je@^B*>1d=z? zLH>7EvEh}#oL$pO1^Ze8s<(~#&_cMI^q=`tNrg&KOxS_Xr9<8e_2t7QOdzi`;oNe1kY+F)FG|3X!2y~Z>Bo>cjRY<(BMI((df+Oa-$?OXc{uKHS zIV~-BEFg(WBh!5%^w5D#vJ)qXg}51=`=k&gwAi-6Gig@(a0Q0Lg?AnmB001%8Q_(u zV5IAdG+QM6`j>bxa8Yct6V9>ebL$o~q=8r`j3rt7G&+cP(%?jF$fVk2Z6Tc6_TJtS z5tp3FBa|~yMeIN{I`(s~-G89Tg?`rLXk#D_o(T@|(e(B8Rd5Ouc>qW?GdRYp&3+J^ zNeM!UT|V3K>^<+qfqPkyfpj;3VH@~caX8U#X=YYLAe}_+=7STi zeTc6JY9)yq%#Q~t6nAUgO$C%!zMoWwce6`a4UD$l-;_=GRGh~UK`0@dbT#0zaFzR4 zE!4n+t-ZUgKpFT=T{cBoiNl;0#gig%J{j`Oh@5XfzaLyoK8zaL2+s%&iyrwTLnW`EJJKHa##o=`{x#)+^^E%lN={`&E0G3WveSB&TjK-em@%p* z(8eTjcZMKRZ7(d=UxI(sEF|j+{$HqvL3p3G)iWozNUf4Pj|u4Kc~(+wFS_ z*F`k9C7awdECed*o!domc{BIUH)fcz%;70n(`X=Z0tyKv?G)s&kTh4VLV#ef@o*WY z)uSIZu`BopX&5-71CAq&brD{wP;T=1Hup#Fj-fmPR%q2Dlcqo>#D)vgUgX7 zA591r5Ci1+EObuJXpsjp@a^g8;TryGR>jE32)Zzx3im1DtLL0}qTNv%e`2svKKO#o zVluyO>ULk*UOQO3D3|==Hiu;L@j3I3BXw|k)^2h1@0Dx5pu0G>xsL!|qBujB$Dy{( z{%9pNhUP)~9Z~ZaUS>CUO=b|3IE2+2Y^nl?WxT#GO7fy8n_Q!EXL{$*~+#_rLwi3qbiFrR}bmbD^Zxp_pB^ zV$}q5JLL~*1JFOwf!21*uJmgCJI@%Xepb3C9s(0^4Kr5`$Grz22Ny{>90Zs#$^4q~ zU9s!el$?+5#)bb{w3L5&2+&sYFCWGKVhlD6FqSE%fN~*cW};Q86r9WWHHqw9QK7bq ze5&E0(4}&@SZds-)hk=MPXBK~{kL~m>R)u|t2q&S2Z!_jdytNEDL{9}9{O#{#08=y zDH7n4V-xmC->_jr1}b?vl^nz)B`rN?&iYqY9z=ZOWcc(yG{wy*`mtER2~iZF6X+V` zr=iS^uu39_-q7e!BC^kkvN)Ipe^Go8>9=RWZ>uCf97sT`YHrRha|H?AO}Z(?lZGK= z5JKwhUCOiZ^&Fp?zM{AkkSOzSFmP#Jw~TtI$kZQqn}(6i_2_w4 zJv}R|Ix=KTjB5mkTiTUD;IqR?n3m(SIPH?)(=fTFxLWuL-`DScr9v8A66L!{u>jx* z58DSxgR)^->yhX!_V)Ilg$=`s@n@8jj@o%Hx)b+92ax=iWxXfOm+V&yANQOc&XCyf zW4rqf+=~c=n3peK!fKGf2pQRCZfU7ZCv-)~R09gL@fcNi^N{)wR5`Gu7*wpwPKJ_} zwR{*T?FHNdz|CH#YU=A9-Ivl$O18GGTRxHux-gzG#RDwoS=n}Ym3hSOod!?wTHcwZ zT~orcf1%}nnJS144Vkw)uhS0{-xPd1x?|kTdlkd&ny(nlX&aW0NIy&CQ|s_XG~%p^1y} z-pCN?>Y&!GX zPoJUbs-Z5)*{f((Jdc%;0rwpw)+r*3G@!s11>Z;%sPI-6Q*FQ&-tN9-Y^CsB=;4%! zyZlbeK5bAP`N{HA-{$Nu0QA4dH@1|N#+Xmt{gk?#Yu8=tGJDTd;NS3IeaR_)Wq{_uFJIh|%aic0 z27USRJ9R}{;jgd!<1w!Uzq|=&r##kZM<3>wuTeFD5ud-_rX6#NHz-vd_mr=UN`o|1QE=cL z2h338whFi>!SZlfrHXLbWu%kwH3-RUjyvAo({zBHOY_q8 z-#6_2U-Y1V8jA#HbN=rAZm=DPWa0K!dF>Q4KMBHL_&C}pb9n>>Z{moP<(EJ?s{KFB zeR(w2ZM*gzrIJ!4l%Xh+G0H4blp%@=q0Cfs#L+pMT!oC-J*~*L9uOc^=1c9%r&1??3;w7j>SEogKs5 zhg$42Gw&1id-d-Bw^e8~Qv5%Bf&X*8j%H?NqG^9C+V`%RPAH*Y8a*BI9$%wOCr;O* zY+r`PK~d4#OtJ0TBhu5Qa$GuxMgz=Sf>Vt{b;#ym2S%Ef!D`(J&=)>z7i55lFsi}+wNM_ z^$cJ)@q37fh&c4U%JUYk(AC`BH3-f=AjM(WCKSAicl68lb31RfYx%H5)oKfv4rgUi z-*@CIiQjts_%R#W?DAZDlzrpBe|N-d;F;^cGCApuKV1n68*}(xKWevO8?#HtdReMx!QW& zRyvj#@?jp26eMIbAd=CSdkO8@XSyqvV*nG(8CQdE@(G!pE{+`4*Q&3DXH*}+=SzS6 zwXSxgbWY)uud17y8~Hw%4d?i#7WIvf8=9TM@Du_D0n^LiZ{@ui83~=yG7t$#kW-_9 z{`RFyi{a?%Bp@i*-XU>L<-fi>8S2Y)53geq;`L)Y*I!#T@4i=yrqSWJN8>YXdfwo; z!m}$;^5~Sdhe()j*!HtHJQiZ+K?K3t6%#^yJ%<6@>dNC7|uhQm``*;r&V{`AY zEA~$9`5fne{5YfI%2{e5_4bhtr;bYpB#4|-#`x&g7)zxL@CQcw=K+YBa454dF~XC6M5)H`kv0b(CZ%))&Yw&Cn*Gwgfe(ir zo$HB@@>Z-~y}A^VKguMudTpM{ebSu|%!DdyjNG5f{Jb39rmNzuF?42RWCV(W;PdAx z5FQ<1U}TKlZmpTYtGa1Z5Kx}&a0qwwsW)7h5&-&S`*}5FW@n6 z+_<5Zed>cZEKp8D>_Gb{BqStl{hH-fc{!1 z?3YrYf7KwPi0IaQmmZ=aG(-#*h-9+!>uB(%#CU}DpFe*Fad`zEhsM4kYqz*4#ceSA ztedbSkf|Qq8-bs=QyqM)%MC98r(Z2Upk2#!?a>^^xGETU!y@-F+(#VS^P|6faX^Q) zHu^`}ebYMXYcFf*>)XSr8=5^X)shRd8a%WCdd^zxSs3VtT)$2)4Pqp?JsOo`(KCH0 z%}gQKDt~IW8IyR>#k`~oc-tk6AYx-vhna(;M0jXuCN*)!HNTN=eo_g5q9r~EkFxjM zq3pd{k$3MNz*}_M)g6E=iE`%m&y%^lI@G!;Gx{}UrsCMz&_-E_1;pdb* z-U~4b8A^g;;xYhzC5}Xyk5xl&lCRSn(VedTQRJ3lz2SscTAufa0DL+uVBxiR>C&dB zhQf#&))QgeMpkGjWJ5@@#8x-UI}?Odu950)pWxlSJa|qM=JVRP8U_XiFn1^jOZEO# zk&MFqyi;o`H5T@9P*4!uaHF7DYs3#tmQ)UXqnx9suFimMB#oPIB^u`Msb6OP-cb~X za~M~yS@HTcnWypt#4EXoA;qkMdbl~uUO%a7d*H>^XGcV@Y7!_vbLux*fNn-a>~?T? z(x+M4A+MJ9+R7zf(6474!0Zj|+D!Q5-~!^q$Px`~3^L!1Tjnp=wzae&*Oqm3IN}H= zWpP*#ZP~iDdYj$8;zjeoHAxhPyi`C=?iMaf6CrbMjA{&8&f3`2#ElCZ9V$$t0p!#X zcOfuGm!OBZ)u2@k*IU_><(N9#6Laj>hhC0?)^mXcfiq645#=$0?_kg*~CHd zOmFp?xT!YY{;#w`e^L7Jew0_D>GiSk&bxw#OXd@xU%rN$1I7!XH8ui5LVuwY1EaNM zVdFgkuZ1xTOryRLIMOjt(^m|+`Ayf9)V0VCYqtg!b`~Wj@`HnVx~Km*8yecKVPb*- z>vi%)An>GLUfJb<0OIJjjJEZPcnsR^`NT;yQ7VcTVlJxy^7_*yJ|!-@@jGBokLVaM2KUN zsZif(sAiGCSfcmkQgE=T@9);g=;%uL{b(FJwjMnSVh7tiWcRH#62G7KWa*1)-Zb#* zut9j!YME|T$qZ-Y29y)_B2PeINF!vTG-|*ewecAQ@(m>fUs+jMA=u%=PYsq-7pz{d z8lZd&V&}e*3l}d!)3RJkOUn!+1tA(~JykTt_!I~B@=z6ThI}`~hW_BJ-@iDKh;Y1S z^=bXRt_9;+g{0_-C#3*7&T)-4GH41K@H@**hWLrpi zV&p+wph%QM%JIY3*eHm~lncRA3CKhMF1$FHbzZBRYN#vG8r0xe~c~bPk=umV~_UHYBJSFdRv4 zB|}Mx56G9+TuJ;BTd&ji$z^EVI^@=T1I9H{Z??^ zmw>Fq5+fj-TVA-ux>M`od|3SDZUv16j)BCuhjqXl%i6DA>2|dB@VqHQX3AS{pLU$T zKq+xujLr&EBd)$i<%HrA_oG%;0!c|pIB@F&0|QfXK-dqbpZTn=!;@B#5D{U?opKJ4 z3vwu~>gf$rFhC=>EgGh`fal5PPJHKJ)A&&9;p4}IZ)BJl(5P8cX&&}t?XND52(}~$ zDp`2s-=4ZRS$|iF9mcyjRnEVhI$(FPs@Htu9rkT%`}Z#cSq#?q?!&_$#uPl+MFQ>e;j^no7}->6M@`Jn;JHlM3%upWB303`yxXAa{`9hwvGWMf6qSP`IjhaX#057-Qi<6w8spz7I%!2xK@soo|4A^zq z-1}VPQ)JqB^))q_u;J;D_hkB%4`}PLu3ANhyIiKS;z=nkZ>V%I-D>C4r%$m5l|1Sb zx^?5c9u4-|4-KCB=F-54BiZ!ggsM)NsI+uuwL+aPwT^1$snok&zv66)8B^LUtpsMX zO9GY>YjSi7Y}&aq8qk;Ay>sXV6`r-$i&o_W;9>dlO5u(|a5^KNll; zeNdX zpHd3%oI3bhiPx@VVnpg;g?%Sm<-L~LURhC*4yfnZ;is<z=U=NKNfn zR^RJnDEl<>CUF)>+rIgMEpGsQ3@+$5==ziQ z8-H#(o@z$-mMAbEy)A5&e?)?@e~zB()XJoWyVP+g1qlvim8#qD*IyZuy@ohu8bu{C ztl#X16SKZ&-$m}S=m#z?PLXlcWu@ayoR74l1r!zUAS`eg--?VZhjnH(Om{JgNQq2L zqD_X~6f8Ux_G9tNcN+z^sL#2SjMl2|;_OxN1;Vv*n^BtM7-KbZZdmp(6uCJqXJL5= z9oqNKlEkZ5S!yR^kRUSU_3dRHx?0JNq7MHNJ~sN=+RSK@Iu|l|^pnGmW3TOx6?$jT zLL~54ti9L5ZDh}WISB2*(!XDvXh0SOBfNExRrWwRmht=X@m@XDi<6KI(BKC6Jg~6XIRRZrkGbumyn=9zo4!&O=EiY`*b9JtQ2u%ZxQ=(mS3N+aV0A zb@*jkSFIw__BDKb^-wv+Z>#%se_>U4M{g%DtV&huYt*VzvARfpKMeT`(MO(!!)ElL zKf>U&`mQ4q;?>d91$dN}!lfS3Q(D1sv8`!ECF1+6F_TwlBqhlP!sI;+i%DQMeYm%VMk5Y@NbwBdc7y6@R)BL)-Fi7^T{kyr zMA1jkO{{XVPN=9n%DU2iOjiCD()3daFOR6nsa`i~?)8-GpYK*S+UG66H@0}ocn7DE zTYgaoxl}R~ALlB3B0001Oy}Xp$;nFWRw331Vt~>o(NM4lS=98za6gnSqK&(vYk3-M zbNBEu4);o*DRx>b%MzlHsPKC-C*@;R@eO^QQxYnHbA>eF$Li|pS3*Pc`^M9@zL~vS zP}R4Rh26LH~!AymX(#Y zLQF?jm$Y9_PER*I*vASxy;o2P=paBWhe@)say(!c8l5GxvU29Y$)^U7jI->n8S-~< zY&+bAikHk~#;Wko04(-Lyo}#va^rxtV2t)F_&LIRbPd$mS!Ra| z_Taupf+mjhSKPc>*_2hmplYO3&{vFBj6}2OBom)9SQWf8@kK9%*VD{s` zOB)z}&ACHzB6|i|WJXn7;qZtX2X`_0{5pFH_o%6aQf2=u z;&gw{lnXWiX=ePLdQ&g++@IFiWGcK25-sPIzyGkkGdA&TUMzL6cb=iVrJ1IZo|T!2 zv!OS;QttHMWTqRg%M#craY6ii0QO)0pJw~U#&8*%+OZ=F7j9&1?8(@WY|J^}3#*)2WS?^5v~(=<)kF5uv0~whu#qd3jyW7a z-)B(hBWRp|_9W1US}b**Tc37PXHKg%U)EFLAY_mzK}m>{DLfG5{n>@P!%*+vma=3J z5fOQL`r~!Th-*{B+ZO>hT8w;!jh$UfN9TdTBkjw@yKpPw0Z0z%{#|JwRZy-iRCv_p z%o!O@(;F+Soj#@r}^ox}Z`KpY)$m#!cu}yEeFkk+HEJTyMF4`WE%}*T)qb8e+fo zw2CCI^O8t4c*G2p7!u(!aJTI21q5V<8Z(Ix(yjd+x z%1Bv}9%EAA-jw8z>U;K#OiUyL5rm^)1j^@8?8mnyfq}KBLq%Pv1?xUL-^te8>wlfv zUj+xzgBI-=&xviS8viNLr}P&}rX0r>_0B}m8VY4yf!5yLhmkh%@=bq;Hm>8B&#_xR zi|=Lr`Mn5-sn1nnjBCFff0e)M*TIBc`pP4VDVUs_OP#}j3`=xQ#*_`fc)cE}1Zk8E z`naBj2?mC%x_*5Y-m6?=BaOjLhYbu2zWklK0qtDwey*IOzPz{jRpJ55CG{BD+1i@e zS@7-h&!ep{wwI6F5Z@P~JOEE0K$_JwTDA`(D#pN1JWUNv{a?krstn=EQ#yWWXBzz{ z()LFBuq|jVwRgVe>Yv%)dtXR#5-!0F#+%-rpd$cxJ94S1t6!k~f~Vt>(7A%eebV-B zErU0!l-C$qQtM2Rc{D0ibTr3dRybpNZ`rH0eZQL{r(UMV4}|#pD)l$s^bW)=E*n=R zo!gFTZt)ov&#_x;magL2nodL6(UM`k6q_){a;GCfIe);C7Vk1fAk)OPPu%OHK${jw zh7koIY~>4Bn`wghex)RM=3iIEeD z<=8AFlnG=k<)giw9Zm#eDGrYr%eye|DsVqlBr5&=oEva<{yzTTa{y?j&15pBiuZdHIWN{qm7Li=5J>2elbfqS{R>ySuaNfIvXua5@KU(Q1?4J_YAmu z#1!-pQ94E5_oqXU2NsDy*GN$7+yot@W19NUd$eU_4)5rZS}Qc8yx3jDIXPzP zq}jDu=#qhilKLh^@+i;`xP)$^ph-yLr{W~il}!IKIs1B^6j>uPpW629WnPV|R|<5$ zM9zE{wTU$Rp4fTqvL|FCfJ{gfL=MRY>0`5Mx* zu#!+4~qh7~~?#D%O5|TFX0x=&qurL9(L&({TSn(JPC;|4O zqSopq?s95DY*sO&x;=WkoKJZWIzv0mC_#HfG*%AAhbsZZR3l|{q)x3@@_hcGCkht- zvOpZVaoGuO-h2tZx>ra}n0p_t*JtwfWnXzg5($6;6duL1)o!k*YLQWo{#Z8P!~5_eH2%)?L57 z1kt^%j!z~6TMulaEJ6#oG3f!%q$d-MF#(I-g4DI|s41t!>fq$I)zP zWo0!9=Rb^*7OAPJA=mEVXNN=3jg!~Lgt`DOm6E8Vt^zGr^!O^}*1Nx;w!J1gCo*wr zR$7(Kv=MiD7|QTyG&;%PK`@+UH&Wy}r1&+})f@2=P#mpukW4#nxb2ogLalSKGCkzB zb98O|k5fgV($1oFXn^(SbO!6eFFG?=mil|$h_hSm#te;%V3;VKdOD{5mC>XVCO zbdXYN!Y01`7XZ47ii)V}=>ja8N%mbdt-H_7_>iInb%WRo_nqTiT^7I?Uc<7!9LYLt zNjIVF?*p(2Y7EZ3I<=A6mr2hE*|q?+@VbM@$Vs)4KLG|nV8uzz2WvpneE3Yk%-jRm zCP>Q#5fBjSaao|BELO)_2lKXo!|ZnTEL1YG;#Tq9f!Se9xMq4#TM*b603j%VQ|_ae zVdg61Gin1c@-;S*xQZpA24IXPHiDdTLN9?jo$c6i6FF5=uFHvyE>@8Mr<#iVyvPER zEcX}jip=z%as$k{2UlSW%+0I9`IF%S4+J6%tfY6(j=iCkh*!^rtfl&mOy`{gX5M`; zb3{ss)}={YR@`87z-ib3#6lz^2`%g?Nb8a}y4~A9#WlCIb%|kb!Dm$plnRdB9%Q-L z9SD8F)z$m)iW%0%Q#<~-HpVapFETFf^Z<8sqNqX6UD1?fAuZD)gk%#WT zem5H~Tsi~(Ur&R2I!&En&xw9HU4Vv%3-~Z_iz&yVh#Yyg7HVoRj+v#mscEl)db)!k z`|=@$6KJ#z;sA*YC;+!(^@8290W~7`S<;2dGWV2AWpr+vdDt1wi#4r`vJ0g3YyW>S-Ewyj(K53lY>4j>8*d%W?;kb`_^y{J?e8CLkSMo zcp>7DiziFMK5*cBELrDwo(5E{uf4ThtuKRG~f ziu%bDa%)4>4C+Yii&kg1U5~0_PBuH5DRL`RUhj{xU;~&X=s>pPViPBX7p^gY?SJ8& zj$wy4!4y>~-~J#B#(+#M1$?eNH`&v%DtWK#qrrF#6TorFz#$r5ES|fiFT>POLX;Y% zqrsD`_4mILgS~tGqY_lX;r;pge0RdpE|V-q%F$QgJ4$hYCaF%=$)^9^Cok~ z88cu{y95(4h1_ixCyaV~2XAcV9~*vcBkiLE)@DtF5-F7N z)h*-Z?vB&5iUOFC0%O*exJ|ImzH-Go`rmJ0V`Iw$IfCC$2I*2ZZ`tAs|7^*9LJR+% z_mA$P0R$o|)1?D!}y5-po0|Pehj0p>E+qm&Of`-+O_28c#11Mg+ zWXT~2*1#pu(9WQesZt8GS!{X|Mn4L^KeysLNI!Tt?yP?3NV9xrF+a&T-C_|Os_z2} zAjIC=w+9J$j=$jXZMY7e7qJ%pvn$Z&$BKcT?^BGz5i`>RW__<}S|_!hDE5!V_co$v z#StQ~BYu_@E^IR(|1nhnD*TaOk~oN~lpM150=X0ZBTE2uuDZJ*{#aXY(fBiG;YBjy za^8!`2{)Qe0s^lx3jy38S2Lc5oP>$^^*k-XF=6xC(uoIUm&^U;c=l+8SK8V^*X>#kYp{aKR7~fL5 zZx&{sokyaunuo{Cq8&E|VQ|~Izz%5b-+vESMU3rKXEIzCGAzrP&`99mK8?DIZNml~ zEYFH-hsYsdPUc(KPX^tyP|X8KW&8mo`dF$TjQ)ZOJDvrz@)!3NlWPbF1-PRf=r8_^ zg2E292g}~b#z!E3DSRYY2BYkMMW2Hev!lq+DjxY8&CSNBdgL*O0;9R<5ce^5q5lDG z2Tx1KNjM;&E!N{O(nc($+vE>R=o3P&UVVs6&c?>(ZNA58i4(<+GJio@l7X;>W~5TU z*9Fa>#joRFTFzSB%~7luDP_?Vi;tn4SP7d=>yqkKX7hq;% zT+hH=a0*BU&ZZjDi4|+T7D0e=lAFyEmDZBb0t&J;v_IFo#S>&NRsX<&M<^G`(!eSs z(bL)+rikYY?>p-0UAOh|_b+^8G=zJK89#>bjd+a&oErZf1}LJQzM5@rZVvU(YqryK zmfOg}RQUW8tM-TFV023DrQ^^R$(|TP1Cw#U9Cq!B_D&RaC=+Bq9(i&6F0wJM>J5)e z*;^T7w7Q+`61Jw{5?_ET$B#TaoCF*|t!(TZ47r-~JwyqO|KwmU&ipT8s{~Co1M=lX&hn3FkZvk#;;Vmu#5(EA)^}L zIc626QtNOPO&m7Vd9oi*P~9#go6P_V014U@@{qwH(ZHaOJbO^9eJ>8a{>11-Lc0fI zyll8Q2}kur<9;LE z8kD#_`{e}#*DU;6<$eF$+#(ru<*dL3-El7CE+5LOtJ5VgDt13cJXqrQ$e!>@*|Z4W z36v*CQ}67{S?V}`bte}KBrrEo4kNE+CJz!NVgBhOmCrz9znWVb}%xNB&)dhe)6Z+)N?yFF1!AAWdKY?b}>& zFVMXl(msd|N}!JnU1<9IQ9XWnyT0x^u(U8<94tmx+BF13Rq<8=@TDoz^j`z;t z_K`IlwLdMweC)3f6N78D{T!>^`M5^$-3PR@$R%K}y0sNHe)EpePBFX4SN*b^1a}&hN zJGeJuQsRNNsi=)_!U{czfKYP#ooP_K5TZF3N}WDVu5Ox0`cM)GNVl73v)K`m);<}So(ID^6p3N7^U?c28>Rj~;YaAEb)I6ZcdiCk>v`DZiyRuQHS;Nazi zeqh7IBq^v5EEBlAHfhZl;)0r2yFi5A2koWSQQ=6xpV~qc3i4+*YCd>J-}Uw53w;DG zs9JKwYUhD)WL}2ot&yHXRB>21iWw51uC6Zi`omg?Ci$uf!e=pVmM1`jKE4e{WZCz0Ha ziOB#cPX5_}SNI$Nm9tFRmi(BQ=toJ^1L5{P+!ILP_>kjZR74pxmH@tpiPn9T`!O9Q zQ&l}XQ5*B)rbecG0XqOdv>Mf|k7Rd51B2K2EN?+v+>!Lpr*La^e`)_TIfs*^^dp2a zCx2M@`oGTZZGq_vxhzF~_wQG|A?yD);^+UV-Tn6||5vT`lli4v{Fprln4i*!qFzfw LZ*Q8q#fAR>NbM2C diff --git a/docs/articles/08-more-epic_files/figure-html/unnamed-chunk-3-1.png b/docs/articles/08-more-epic_files/figure-html/unnamed-chunk-3-1.png index 4d7f2f645d9a4948e32eb2fd79b4f361a4ddbe5d..8ba6ef6985e4de4925ec931736d0749985558bc0 100644 GIT binary patch literal 32544 zcmeFa2UL`4w>4O5t8JT_R#8Er)iR5;k|f(LilqW3a#B#qAUTI_TSZA)h)7mZauBe{ zL0f`=fRZyv&PC2)_EUb}cl)0;|HOOm%&dERtzIBi)cd|qIA@=|_j!15QC52W4;(*G zD3tZIvr?BSl$ACV%8KjXufk8%mse=u%POt&(o*<}|7K2QIOEHj+h>){DU_Y}$^Xl| zI!A6&C_htZQomlc2^?s$ekNa)CAqkf@6(COwZC3`ayIwg?!0@7YtDooqbj&_@c*{? zZp8PGZFrWYJ-np-@)GY)oaYSRpK+%dDzQD-`|w)sA3R#8^OoJ$tdP6Yx{>Yv zPAesgp+;d9hW-}Op;YZa%WCD3Pkd!kbn5z{u{xa)54JsLq^5?NLQlJ~e|df0_r&6K zdwa7z)8pa8hiAMjzy5XCv|y)o?(o+S+9^#|Zj*xz(r)ZLFU`A4Oj_?xm1*y`AovmE=DaavDyi52$7I(%CI)6R8p8L)?B1dG8&9CV&ydr zPn_}M3d)X4(kV0U`+73VeW$3PV@qkEa8l&`6&UvICG`AP5BTdh$%(P8=61O4Fp{@b zAi1|9T&~f&#qmp;dAG=%YPyBqx*aEe;c^&$&(UC**7ATdAuVdM_?{PSk*LV+6}A3fvOyS#bpR-wLW z|EmfL`GYAfT;aiX`VR9W1(wMzdf`EB_c)DRHL~og>I7-@)WPJ2wyinSr=L81T3wg@ z>J{~9&iv>$K|`g@o3AdczqL3&GZMsf*q^o36<~5E#ki?&AVnl&DBa+bzrf+$yN5Ez zN>AVhIIA{tKD`urCU1+P8XFs1(yfw*LDtP_52U}d6zz{mYO~InQtTD<+khJ*PD@+L>^p zkba`~m3N!JLEMAwe7B?m1dXcO{G#e)gqk*+WY63Z!veNkRX8y*(;ZUaUK9a++;Ec;txtij87#-aC>V{5kih?4R7W^2$DRbMtmRyL)$9UsN<5!<8v`^XARM z!u=tE$-Q++IzEDFr6TEvk6Y|&#$9pRb9D>iCKR>1s>U!2VYoeNC zE@vpcf3$uGQ&ZUPB^m6nIMYqDwY8;FX_qxLf*odi!wW;bB>V@G$|P(D<%Bz*=+6a- zSnAh@sb^TxPM%<)+X;_-;;Wb%bl=YBZ+8OY zWxs)x z7AL(1=f$xQf#A7D^D>!+E$r-qJ>0dJERFJP$GP$8V&0t0{N%J{CYZ85M^D_QsJQW1 ziBSBMsldU*htHEio2<)mUg8KT5C}3a749`%xpHM8qQXU^B_7q}3k^Z`LunIzb1&VF z7Ol07?zgW?yh;1@*I(prO+@m+aE8NQYPFTc;Mi{$Ygcw($P|Jr}=69 zhLQjw-cs?Y#_>f=X65tQ`vE4|w)5@WOJ|v0+xhlk{q#9HIJ$F*i!?bcElg=6VjEPo zAP&)wKC4DRw;6uF+Q5OU#-dcX`_+W?;lqd9E3e4KmUngfYI@@JGg4CBr9u7bXTG}g zmAVF5)gR1Sn#x(qIu}5v)3-@D*w`(l_s2!!`AYow_Dd{`SGc;?+V#h3BP5%A9w95+ zaNBp8n!jUlZ4)w9+hLl^_62*Xr~LK#*g@44;~-;RmA4V~f=H1@>FPCqDT&*SyOKO2 zm3wES#J2NKk3^PtJ-H%FOIf{P$jT=}`QA?iU<;$1ToxIN5ts70HGe?UK zTPBwUt;v`h%22G!vFI(g>5om>tySgTlFlw`P#vR87GCcH@FSp+{}f=Ht`|C&zO*+sGX}6dx1gUM4YjS@g134VyJ_^yb}YZP-&b?FwtI>mO>VMkAR(K)X@hXbqtw~H=!&`` z#&$8A+jc{?V_jACR@gIo4KjjtuZkA=N>A)*%F)Vu=&Cf4#wnnG)uCIokf`Kn-3H=*XsEA_zO}B z>uxYtrHIuk+^4Pt^3~{sII;^h+i0)bDZ-!J;xwmQ6(#>xzv@uBHIqwnuCA=@sj6}0 zNI`fN$E+=vsFh*8Q`aXyTgQbV>zwmb0^+bI`J6m6)tP{jgaICt4iLqr^FjLz-EuB0Y zPdBdGutO^H|1LU&rG+QRXweg*QIqCWO2|Dovs4t3+Q!sJ@Xs z;-kqg-52QcZW6)P&9)PEi|5as(;tj}5z<}HuAG78)-Mp&XG{`|PXYC`R7w(&ge$LlnyIkI&)q)pW8|InQ|W<`*uf>95#5ape+I2^-&t#wT4M0 z<%2+xLMAOO7)Oga#0TTgwKd^XnpS6#&&c8&vw*6kf2?VzjCYEk8EtxT(&pnrwb%KE94}_Cu_C6YIQ?Bs%Ljd?=Y1Ly#3c2 zK#-c8MS<-3(c;XXxT^|-UIeMuq$ZGkGTp><@@;Mbcxiv8?fJlJus#nTveSE?%)^cM6{SoY(+k03!Y1=uH62U?P4l<+E|dm|7K{{d za|2-;9Z^gDg@h0}iJ2QxQh`X2BcD%g%+vM~XGh96SV-;fN3r34I4%_-ZW4+41SaAo zpzIJHG_^l2WB2YhQf#POHwgqI)Gqz{9R*U15$A=A~M}U~L>)+}f zv+S#Mbxj>B6>Wd*tML%gyxXyxKzf9~H?FSFe!2de1Jk}GdP2>4`U7|Q;I32m`1LcL zku0?bY7^KJ`hK_zh$2D;j|7Y(N^gD!sEM=biws9hk10hkn_gf#w?vo={`C$!=TRvmSN*zh|WR88eMJFRN3CrmZlOV3RXwvUxbdI4&OGaEXKeZN6g z;F3_JPKJcZ#Ee)?*P~M#9xxFe1%`M5DUqTO+1&ATuO`Y623j+t5ANO_CKnQ{crOB3 z;Fw|Un_5(Dp@4cr2wCI8q-Y>jK^!ugF0xiuKGyQEk3pNu4Cf%0LabH(XIas~Cny<2 z1-PuInykZ(%1*Sq+XWrAzBP=Yc$5-dhlcbI=VRweN?YqI`u8WyKF0;J#!VhN|C z=N``!y9dWGxbcj^KUS_SPX@M`Q7OA%kJ6%I@rSi*y#c@8oC`p0y4RkWW$ja-S+lO$ z^(-bpvpp@ut}ik(dDDK`3x^IJih64gL^^hBI@qk}VAW(w_O0PKCof49r1JGyi03qb zoC>SZGagYBnG-&ucZ7D+1&!*dW#Ut6K2|1K6AFnlT>;^P`R?1jQ=CaK49TmPuup5z znaE1KCX?1}P!T4>;N$hC(zp;~GYe9>Qmebky(HJ&3OJ-49TRf_Nm9Pk(Fp~MqMvZK zVO`=YR6cqEl8Yl=oVmhGAZH6_+=aK7#B4-;cRTi(ERsTS@=ky(#vsS2tt`D>XSiB% zXp9s;hZNtF+Q*jJYucVKFPgOEA=G>}-J`!1-Rs(F%qCI#-KSH1y*ikcrZuCV6y64<0cWSCd(RDIZskDr-f%D zM^+l&foSsvqmiTN;_uTOf*q8J8ClO!TyNL$Sh{VlWy#4W$AnR( zFcR;)a5ziy`d@!?^xuv1Km{FD?_Wy-m>4RF-UU)eA{{4H zDb5T1nnL=4V_hZZCFivB-6Kbm(y$8)X#0aUJI;fJ-y1mhtM z9c7x3*B@-!PfEY7#ZR6*$>m;}E*N(*UOyft*8S|5cKKlBmGCp+vO!Th&lh;^Y|rO3 zG5kDjFees&7)#WB&8B@)xlzmsjOf+0AG(mHIXw3FBhpqBna! zzt9sm!W6Ifa7kGMERdIK+TIoxVEhoD@yge|6foTLM4M}^<5W|$dhR~!gwR0(N7ey}`9NP8&1;h;1{;L3DrYs9 zW)&R}v8$5L)2U3eo<^6q$21%#s;0q_Y+_?0Fn-yJRh6~L5fKOW$@o`3UnEsp&2;^z zAW@-a+wLG&CHbdc5j{`7wJc zIg%v)JrlRLA}K9S38^1pLAQ*D5Tr+D?z}I_eh_~jhbrZD-+u^SkEXnxT2!M z0g$0nf6H?Ks1q`nqVQ3nN}j`qqifR4PL(v)oiiv^iz-|98nyAb0`9T>Li=nF z3VoF0gRzN;$DbZm=U`@z$u`=wdwdC#32=2Kr3)D~J@LYODoOczjv*fyC|NhR%r#H7 z)f>cn>^=W97kO@HO0%5-3EH-NF?)6|A|VzaH*6*VAZURd+>@WJVi5jyY`qrAyNA9? zpAw(#3GKB~UuGbci|F{}xr>CJKn(wyr(l+Y6mm(mH zDJ&nn&t<0@6zn`b{M9nKFgD8((YRxv{d!`FDiT&ijbnBgsQ9G7U9~WTwF?{=j zEG~!kf8TxY9wM?E_JGi>PtP7(`}}$(I(oy%z5qkI@f5OYzExVtjDWMQ!9#QQEnD*b zT(e0Fv3&%eQzM)qIhAv2!q!nAn==-5LO!91#uxW^iCs35X>F|QBS7>muzhK^^Qd>o ztA^+g*T3yY>lI1BBOMESEXaqU2sm8q!Oym3Ad~0AGVA<$j3I0WAVt5%V z%}e|Rz?Z7zrsOOPZaaSCEJ?ATSX2=V{FKNt2rW$$IiO+WsP>bl?;n5s(NoVbAc(^l zBXEF+#{(FI?`T>FJ+o$t)aRI=tkT6*D3pzBHvgcBjn+Ha8y=F)9;(@iud*rjQ$y-H zb6q6?((xKueNE0rYZE)2_ASokEM;gcSPs?;uvmhe#A`VcUcyre1WKD|g5L{GO--$C z=v(=HbP`ZZwe4c#FNPB@)uROrs*jNB29eIst7@NIaG-IBMLEqfIDQ|gjFL)44reU^ z1LaL+Pi2u>X8=!VTDaFYH)m;qc_?W(=bR;vq2Jt|GebJIG7~eGULGt<3j@+3l+&j` zVZNhTxteypuPy9RUYG)q)Yuj;96LCbQ-TVJ z^QUFZRT02Prj4crucvC(O`Gvn#rB%b3v`xR4b~=TH)o?1^`NHHTXwY))2lb z(t3$e5*GmCJ3UC$K0(mnX&x!9vkDRg`}nS;gCg&pY|C>SyOrnb>#L|5o*w646#w!EF_#quwW6k;|Z|thuk!q`H-#@W`FqIy|$W$@%brtGzFyKl`Ui+y^(J{DxVq`wHI6G$|*5q}k z7ApHkTP?&v3ia-t)h_TnZdcHFF}>1a8C2bZr0Uz)dGsPRXFbVwME zKWckQV0=corNvanwiG2BYQQhaMh&&Vq1s@}NiA|na+?wMvYt`>tV+XFo#X)5mdaC1I&KwZ-aK_GI9D{#_uXwx(ZTNv-i^Y*h8FVd+kO<@@%9sLflpbXGk!`~;Zk8tN%>s3};?)-3eYoAbO`{Kn8BrU`yh z
                      {(QreOmGa^e=U_ZjzI5qQ^R|AzVw(`hNt#Zi2C5X>>$!w{wGd4S{80Zft0O)8 zwRqp$9VvAM-#*ufr6ZF9Is?^$p>N`T4&T+ev+qjCIAHvR6j6(wM_Yf$K_(I?eWvLc zZB(u0ce-XvHv(i6t#8Xhs?#<{$pkxAE!a6 za3XseXL-&5i+CGP^?cc5q6thmkXU?_jV-}pW>jP=Cs2#Y*DwLz7tQ+7v7_u=X)F>T zKkOnZ7e%=fLf@OXIxWn;zTWfK-L*iB+t&}ZOD@gGHScZ9na<-aA4DJ}yV4LgA;U$Y-xn1Td@Ti<%0U z6%n3-sugG)uN<1bFYT)tDuw7HzdYmI5W68&|FZh_JN_}j*OdZDK4@yjK=n0QJ7xmB zS|-~LGHS{P{fn#}rZQS&TJ958^Xk>Bx!6nPrx!M$KxAm7#_x;Pd7^Rq{mK!PTLl2) zkE}c2b1x3x&m1;#%zCP+14Yk6a{lYa-UZ}Y0RkULer=iW5;Cp4Bz+pJc{z8jzuDIe zAikR3Ibby_qRZIv4Ah`n!OFS^^8RG?8k?NuC_0Xrw&hNP!`TknMmdZ9`y%xp`&qTX zLRymO(ukLHWql2n*p3kAMO&#DmFo*Ri~TuKZ__Y&6@zZP*{Mapo#%X5^`cXs2tuK= z2M-;3NHoJNNq|#QiD!Aq(W#JXo;J7K^i?0dB>`4LAU>`atakJB5%Bd#;wMM_cR9XBDemybUm9)7`(Zroj} zqRAr^sJ37mNpsYfk53kHjVdV-Y6&#%uFg>3N{VWchK_6}fs((BQ0Wp9_Mxl%i(XXn zA&Sj6CbAi(09*N=Uz`HM=VCk+>upRk*Vs-kFE7u0XUZDxv`$-s^zN)$AI0F$g%UO^ zzb^VASS5)dakYyVe=HLlyO3qz(!F@V5~WtWyxvBFKpwbE9sUe*eVqF_l%%-J+?Fy^ zEzZsyZ`j${^LO3farw2((9KDmAWH?0q)IxqaIJ>ZSU}6w24kR0PjE?LQ*#jcTI=5e zlRW$qDlOFW>>k;I&Fw=#T|{8jw{S)+=6X~;9SBAFDM4siD?XwG%>z85H|GExj81Od zytzYNP{L}UHnr!aTi=!;UW_-1UvCEi6|Z(Cj{95mSMvnDwki>*kpdH>;+rTRt&jj{ zUlkoP+k-2M#NmPkpZxd?3Nb0K#n(VOykPW;5gCEHmPWaKw0e+)v0vgX5R2h%4E1 zL#o;hf3S=6{nHD8!ds#VKh8k8H{|Ln&x~a{O=TE}SEE+S3t1cww}F1sF6TUZCDzDX zXW;3RCmlYGN}BPVHOj`%vynBc>XS(+ZM~p)DP$7$K=IrFa8&F$nVEW`wOiq)TwRqC zZ_u)eE?ju_%PypP_7$tvo!uJs9feZcN_E(~I9rpmkL__YmtD^b{e~ty1WQRWE_!lM zU{V-y@of`;^kXQ$U4y*5ymGQ4y|3O9mj4kpc4e(5hSUs|NSB8>s>MQpLiK*_IUr^F zNPEG4)OxlBXQVEpDtNts)cj0BiqhPx9NpG0|eA*H~JyXCo4AJNNcb0%OIM!^nV z)k^|ji*3D2zL_)4%b9^+4x7`o`FdBSYyrr!bHW;|gZx~27Gi86>R|t5Q6RJEL1qgC zkRWC}>J};j+HH*ts7i`@Q_-3WhvG{F8w|an*slitgP>xo9e`C5>y?&`UFT=LXxx;( zcP;yoh{U{xjC=IySv*gA>a`pbF-+-WWeH9yQm%1N3IY9Q-U&Izc8*lgnM;Dz-B~I@*fcyQja$_%E~gGfraser3Kq0t33)nNfCO#N2A&)z!Po3Rn8k!W~0KYD&>kTg!F##^(`~? z4I(|y4;(n)ibR`9U!Ru5TjN8(5P=`55J$98VI&|L5Uax`$%S6IP)N%oz}C6MG4cBK z>o3T4mgz1nAgXr)0?-W3gn(~1B^Z*}V!)wz$Vo2nlRcdeMe`X~Qt{D@4xdvWtB@Fp znwNgN0xpZD6FtxGXtlokJ>kn4P8LNj)MRl54wmDD=^(5CECw%2O_L4lj%zqCOb`J< z3$vF1g-aS?5A3h`hT z&aZoM2k1)6Ng>4QZTOeQU_GLqagaL$3JmKJmKK>y$c~^ewq@B(Jk<6S-bbbatBf@U zU>%LhZCtNpV8$ z^nhf-1qrc#I#kwiONf#B2IL|ZP#Z8zS#$^y$TRC(NJaQ0O!iq0jd&cu(J6>l2uxK2 z{&Wy(Suq52SV+=s2PFOduQ<=04bD5O}>o z2`REDz(}DcRxU)M5D2cHWk@oEHL0s)b6kHimdt))I0@x0gElxi+W^&rqe)Sn+ysC? zPIW;AGIa-5#K*aY(az%N@v6_;ooR{sRu08adHNk><-{i{^>wy)VO3vBEa0-QtStQ?g-HoN*s()rU#YH9SjeoAJND> z0k6iD@*Jr9;pd$D^;a<=iIzt!X~bl&=MqaTO1o7W=m+*HvQ*pc;X^t`FtRiOl^_3J z0h^rP2)zN;Ef@F3F4TDY)Q;QlmJvEx8exKvIHurk^WitzP9opbZ|9-ldID1%2zH#T zBQ~Cw2poi;L@{l;1L4{faVH;4IcI}mlFXg_dZdJCQCaHuUAtDa(2aSYFE=0cXjw1QgqUZB2B03Ql1F`NTfYRA^OiL<=?+#}aIB8QL zs{$}tY^Yj1u79|f7nd=_K~^nz7+sX}MBF!ps!#y*)((@0mxSYNWi*q78(f;mD}*Fw znI3TKELI0I8wx#;1g#aTH@K5ZjqGI`1UeEUJ)Fe$NPB+DaE7`plpdeX4|XNrb0;Ua z!|<HUS@%dwQ$-X5>m&QVTFj<1{%5$pEEBI2Z!VHi-LVoWVoBG-=5; zeZO*RE~K2c@d~;4sTtK|LqDMkYxSHILadN#jJU>3tK=2)maX3Lqk6gpYpj6;h%o`F zZxivSkvLd?de5HfGJ_dQQe8}gF4r3EC}bG~7qXd^kymzDeE#(qG=OPv673HU$GJky z$ptr2p}qgYv(JC-F-b+)mIoxsO~}-BoPtLnx!pg2G6H2;4~kb(!vY{z&?Z}QS@;=tn* zQScHVJE>Tkm+K%Ly;Ce+r}6v}A&>|sOM({xT_O#!Jb zm5C+38T^GRAvSxDk;R6^8?{DS#?i;uw`N*JCRl9dFc->m%Eg__ct7hBcC9Yb1*SRBK* zU(S3ZDh}}SnzXSkqZmFT5C{MtYCkGn!4X|bwjcQs5|@XF!n=%M47xNn?v;%zS~L6>#V@Mu(6SZBUKhGUNiNX^~P$yc=>l=6o|C2@4~Ym{}IXw$q`#X3}x6*Fs3_h zIfYuBLQ&GmR&a|U1hFZ~HfcK5v|cYf4_0gigE^3nceiS|$_9y?Mq1B#d5JYvF`Syw z+cQxR9Yfli{^I4dnb_CGpc*rhmvCc9FS|v6QGgxn#UMi;b6%X+t%*|~KmtICbn-o` z1K`d3S(TF3CQw&xNa!9Og7py^gM`e0PKVI01tNefL=xdZ8j(DpjY>b^+WS8ygjplv za%K)6U=)E@d!Tp_OPs9ONG`GPxspnU<&r^kGY_=w@;ip|7O5YlCo%KI!HU6AHj)&w za0KOEsz$zCrRb;Cln7Y_Mu`0Bk@uHns`kl>G-0S zwLkpO29%{o6kH9I=>*(F8*ODH*6e4n{CaesK;jF1_3B2Dho1w2k&O6vl%GGq#IumY zadJb;1By&2=KD*5Q|vkZ7z2WnTwMvZD@R#a{b&dXkjh=sj7|~RDQZQG zdWrc9#7O`*NbFcbLJq9P((U0qRxSJjdXfHE^Be*yI4jEYP`m{qjq9*Rpo3@yV;`tO7^=+xe1rW8V? zVE+Zj(DVF6kt6)p^!!*^bkhvCC6DxiuljvC(-6WlA?wv z>w(r6;$?9R%AVaUi$5W87wj=IYN+!^P*uXk4E5nWX~13uL}^(C*YxsA!`abqVI)74VvQ(asix<2%j z+&5V8GHmzb-Lyr8S6`kVw+cP($55#`qG*ufm#Fu`->2z9(7=7FzvMGBb0t0gCk9$7 zMNtC&FJ5)N+{$a#xm;sNeC~d`y82O9wQR2JZZR&t61CDP^X%@XJYi^Pydg=quRlGa zOowNt1D4%FLa-6f-isGsjRLokJbJC6%CL2CXsA>ud&J{ya&p+^NJCR~dwY8*-a+Um zfKq;JJDOgaM_#)(bQCgx|XVFg41LTVDHe9@7KMYe5v&0@2H6x8BSOE9ZAo*7Jj{p_zrKzZ@ zrr3DO1PFG5n`>zs2@4D37ZfxeyO1fcFk4kwc7WoNq6~FGEqa+HbAtQGkT~0O2mM3C z+44q4Ml|~3g8h;pp$}=JfbGU3JBVNbDj_+qo%C-I`c2|OVRh!D{dbf%;$YYP;ZkSR zYf}dJGxsmQDy^f_Z6%pvY~t*&usS?zG2AD9N@aSmSco;z&ktBSu4V?*z16!yH?-t! z!Q*#2!kfU8X?>;$GY1C?@tNRgzq28il)QF4jdF&SQOoKuMNUD(h)F==2bLBVsR}Fl zn$dr0x41HTO3I~g8>$j=!zsFF>0n>?7cSjxTw-E%ouv=JL1NOtLXnh)sJInOv@r!! zQ8?(>i7#TEP}x}R$}T}y)K@nX@}HInDz(JZsP@e+C=HrKhX)s%(-g`=!T+4J;UlO) zbGX9N>{;#8wR%++t>W--nL(w=>P_$J>LP_9>=q83VXVbjd#ulHQv#`j2}<6uXULyhadFV`NoTRT*cXsAW#jPXAA`f?>q8~9Hk{L#CH=H_5%bgz?=_-I6qpxRChHG%&p`pbN_Mi$Vcz&(M`CYmJQcdCv?UETL&O3aZN8Y0Co4nSG|u$ z8Z~z3e|tdM-S3()=e)Gzh$KQ{i}lSJu->SFQT#kM4UW*83TfpkTRXWsO69BHmNU_q z>rI83m}xuE>HC{@I_C4+xd57_P;C&tP^h=O$qo(OLpu(R;_nwC-Zt7OQ5CuqTF|8> zu8H=bLKCGN82qV)qG%-h`@7-(<)e+F6IvAW6nXW~OukU|W9oK@4kVrgGq8bYiWIZy zRN;1azK7dbk@@+)EkD^1;c-6=!e<}^qZ3TqGJ7*T6P1BQWiXJBxi?yxl1QU?3nzb-rB zJhrz%2aPqhsNl`v2)F@Z$PXSf!XzM+PR0@a92&;gr^$WD!u&T!E17V!eV)o3>k3yarEk|p)XT){v?BVJp7JnpSibg zI~~H6AbP<2J-Nc8>g(&vtE%J&kLeTfFcD;nzSlC!lS& zOMMHP$a3<*tDZ}&N2ecXuKKucc04(JL_P!nz}fI=YxeM6p8O`NAC4+0m0X&)u(0TW zR&hY<+gavu4_6~bGF_47x6ENAjMjkpD;+JF9aHR}eTvOoKM#phgbH1?#j4hy;X=(pu2uP@U^`zF`1_Rr?fNAE;I@lk08 z1_o^5tf2Dq%9YLFS$SzBX0w~xvEF$3VyierS8{GZCp!NN;XZb=fq{hVn3^0{?_>s} z6JSe27LsrvT;9S^Px#O}F_5#gsHd-g?$oId>>;8zi=OO0!x|i_qyZP&UUu618b)XmC5h!}@ppm9wooF^+8Etchy1Y@X!d z7)W3};2Z7DTs95$^~X(`Gc(~Lm%DQ13VOnQ;S|<^3OLgxl;ej59M{9f`Y8R&FMgkb zvA>L1rih?JdU`7A>inp6Z?2yrfZTaD+_?ym&kToPY!)U*$WTto9J;ZwnVRq9qb%q zVC{~>sVpL_){aYZBh1Y>tU1*~z zhy?$-<=L&K<)rtMSXRlIE=SbU`N2OHk(NVXk(=O9i~FjgE6U4BaF`O)0q4IRt>S;X zs0rO+wjxAUTU%QKpaMVvLS!TQin8LeU`rmwtRzQV`Zn&DG47cJgXkFhe4XBJ{0+&Eafg@qWv$A=K9zP~#5 zJe@pAKx`TsCyt|Oh8myW3IGrm9Ubjz`p;+o-5MvxkV1i^JpwoU0E2HCcp}HE!Ip2^ z_F#W`W#tnz9*6-*9Qk{WLeYlR*Y&9oqWzmH1~J=pl?DY+XOJ2+|Ayk`Aja>fAq*zU z^urR@K`n&nq57{kyxFLHo__8%UWAG18MSv{;E35jGU4U3t&V^vA=EUC4>F;F^*ont z$K!u36hKU}*ytTFmurcLh$N>c1E*?E|FxA)<(~a~|6iDfKqi9$SfZ#*x^n7 zofT&fgUE4&E-`35ddM&T1Atm^JJAcjJ{>9S1~K5lwK7_2oy|we405^<%IZDQit(jT zvO=(Ey=#E`Dk>@va_HzS5CWF%kUd&*36?|4j@tP7!6*I!0Ts2iN(9_I z+I1>Qd^zQ!R!K<-xfY{F_0!v^D{ocUHf8KYql90OAJhWF@dbBJAE|rF}{RfSWzgG{F08RXlS2^=Jc5;1(9C5^p`jN}ZoG5HE>2 zmr87#DOmCS!Kf?#4(apr^QlM?HG;9so}ROGej%Z)e~ZeV+>#Qs(TtV_jpMzR%Jdo#qC~8iLmP+1vqsVl=p3Abo+TLw3_fkseQwR{6h; zqzRoIB}I@S4igHdDyNyRFDJS!5D`cztC^zq9x-?9{K!A)!iw!ASM3bGfz zK}B*R5zI(Ag`^jO9HT=HN060V7{3rCA^;}{j~NRXx%gkzZI#i;IyD9VP$_(p*a)mo z1o%t-rqn-ITFye-CuxJjJRH>ssQ?*Mc7~w(mC>LEAL|n&w26WDqHv4u6#oVtN*Z?k zZcU&x4mcSrF>zvr31a0gscys;CJyNI^mIF}3}E|mY}&hPw|1c%Xk-ih7VkYIRyTWa z3f=na&o$zgtnu+!uAtWZVWyqRZ6f*|Hs%ZzxsbTHIPypYS)l=XG24^1!Y&HUX0B$U zSg>y!62-Oxc-Ki1Adp@}RMeYF7@1)7*AVXu;c$+_`$;X_t`qnriFN-jLN(#t@wDTN<2PqA{QF>;&Bom- zb%Ed>Dsh^U?%MJGl$l;qs|UwRPDIKXSmZBUshdNI-%BGl)5(8Tb+~WBL9!sqxLsy+^kLYy&e6e| zYP)p{!(U6MLh$b?{zfcVzLxvl8dV654vXTr-8_-QQ3rWbaq>ldo9!W=bM#jU2?@OI z-=@V!VrBE!5BK-eG|+gc;gw@`--Mny*TigYvkkUsdq6G3Qm^^@0~m07CZmf;!%Gv3 zl12wr>4Uw#tPRjXV5>K^ zV2Aky&3}N*!;U{>lI%vOBFPtH!Yit(e5k$PkKqYyOd0}l%bb`2r^`K8TVUtph!pqt zvC05^xB;=4M(3VyT}?Ko#OxQZTcFBHNtr}VLonwNJR#Oy$Tvz@05nzc0N>T3k;OG#-1$ISTrEUDJ36g8V_zQHcc z^$>J;G^pJGsYA>kBqCE+@3gpxW|6mzXyGPnt31A?iU4vGZ0@`NhP7Yc zq)CtxZT zzhr%iOW%+Gas1tX0!8@WmmvG}-`Lnb&B0|+u}DzBv5kHH#F4r%SF@19axByqRyr~q z$kBf+Mu)Y-n-jqtlO9-l2kR`O=CR;Ia;%8@X0Xp`V>1+&{WNk&%ON?|Y8tMtv}}#N zdPo`OAyTK|#FeY|-vDdMtGEscGr=;k3iBEn+ z*<}VFayZI9T6t4b5P8%ZO=rjK@z=baq^*vFlha;Y8r46xZc<)NYU>|skK;hQ-@5jb zz|4@QP|g{mm0&wLX{3M>5mmF~Si_PTEL(%n7qd?`Ftx57iKrNS(HJX{E=P8teaYd22zl8hh=!NR@qx6_Q0|2Tn7a@=E@od}5n=uaok+S> zVCq(lg<%u_B}c?fweA$RvyhldE4zsTmb93fzgxEY2GW*o&x`wsz|}j=rMtH#KF-99mDH`YEV*sGZ;;sDjSCUt68{Z9 zO%>llzWKLZ^B<^T{C9x*uQ1e+i&b2B&I8S)*9lZ>0wht<7l6&ExfPj9FsP{egfFu? zL&^r!MbYA`RDZ^y#wT}@!w(3MVv*`lp6`6u9Q2HbgNzyF{}s*cf3{x96F#pN5E6p$ z@}ge5G6oN)2o?BwE}>4;t?FLaYiIJzPFeSEq!@;vBu`Jf74;OvMpK!kkK&u``L5wE z2j^uj*-^gz5A0@^)=)NoVC0sbZ{<;fMux~0C4Q!)9^B-&CS(=wz^Bqh28 zT}Jvo_nIzj-gN`#G!Y`(9Mai7ng&QvHFU=+qjB3)q%R`4fR#e6_E~1FW)hG<3^~9? zPN#B>^!1C#3A@ypuN$48K{NWdRKJ1=>5xvvSr0B!KJq*+M)EcG!xA`IoJI$8IAQ02!%u1S>gs9~ia2FW{6(q|IV+Y3-uDmI z`8zH2&FXBXoa7c@v4OLx)XDen!Kzgj3tVB0cdEq3IqN{%q#Mpieg=l6(nxO+PcUg| zE5o5+$u{$VH=ll8y|a>hT8)zzK#R1wx%nCQZI=d*HM1{XUPUpC|7VqPeyCXJnbuC6q&0&+B8|ag6c;)N01rdjrh(vESeD`;ot|7Xv$aw*C4GHn zgBI8mT4EIsJ6K6`h|eY?)9BqICI*59IiL$o!WE4~G4Q9xQdyjkQT-E8A&2C?8;I%a zR58Not5safYH6U~CK{)&@-iR1Q|(@Op_Hs!y8%Tak%%j*tDoXdLF^kPhatiiR`cGI zM(cOH(?qnYSTkOooSc*>3x2fg9EHLaNd88^+IOma(7J=)Wnj>8?%9i?P0J_)ff$7^ z=1m>UcuEzq{0}-AtG8%op|qyXN$^x3BX)HYFXU$v?!Hzh^NQ_2W+rc@IR`=cC z&h!@zhb@LMRdiAa2kdM>1|a8Yqd6@mPQrn;HV%0GA6jSR0(P+8y7~2s%aK>3#H9on z$;n-iH_sxukXo3ibu+La1bBLSx`QPo=l+H)H&Im$kTYH#@dwG&X=%c=QPe{4313>&Hz{alZXJd3Dy9EE@%0gL71D^EdPeB(uYdm}BJlZ=L#)W*b&V@8{oJAhKuU}U zkm-1ECQz>}slGnWCL3FHo?n<%w5L$?HBX&7MM!xCdw2hCa;36=zW$wt3x_p2 zjJpF`=5N7Q=L;sV8$>h?!9Pzj1L6INje}U)G(zS;q<85^&MD(>96Tl6n#AoUd*bV# zt>pA{>^!W=I^>_Y!Vt(hI6a$ge^^SkW?qu7nprB) zblGv~)NWo^y{F4HM(H(ryIvf)`BYg?PhYPuLO@^OP}t%9ydGaazR`bh{gD6X6I9!r z&Xv5nkG{Ji!n0+i(dC1@HbN^qZaX;-i;8*%-ueE(@WhN#bd@P{VP>FAc=4-!PK|wz zWCHq#uH3lsDl;p~$KO8+M|K!Meee4cdSB4E5e+;CXyxG1ClK6RUFd)87J;Up-_LTmZL*!kFsaGLd%t>+~)lZs&LUOY4X&0S=nlwzs*1s z^~~&S(T5MG&{7i%EifMEc13OGOS|p2Tv-q9qkuz1(i3~eutbAip%;G z&qBS50P~EVELnPGg0^&LSC>9ce&&#HNIlNSH|P}7)1!-1rsX_11THi5s|+=QL~Vwj zBgot`w7?N6t686zdiw>rYGB}w`~8I_B~`E};H88NQCU!DaSA>D35|FJOB!hJ$?5lit40{Ls)H^aC|h28Bmm~V0?r6H2eI+Dm< zQ)xKoqR3FNACE0qP;x72!;TZz=9d&it%sgKNJdWhIZ`)X*~G+T&q2fEejQ9^Ji2Kc zK0SW)=sBKJ7d>+N)iHi(SBOR1T3?WL^B1I=;z7}7nwn*b_Mr!rbZufTB+9o^mgBY!@r!O-{7 z($eC`5#y)NoGB_SjJ(jVl6d2RnVA{C{iGp7uOk0MFm7ge_%>eG2l0J6-w|7S7SFDi zBpCVpB{@VI@4l?78-YKFb2Cu$t?pxJW3Pbh402uBPIvJKl=ur8`D1oq^ZGuZ*^X|} zp1=O?pMR3kMKNdxEa{PEJy81st+jd}+tBpE=j2=No-m6ASH&<;lnFfX#iq4B(>4jM zj1h>HKE=$9ly?g+(YZ!{RatozMu;oa9ukNZE+`s$Lx^(!M@;F1kjuPH_IgYDD(Ztn80yorh$wx`j9 ze=tS;Zfw;Dw3j6db8Xue0j~3)q-4ub*j^_mr_EAqDIs0lTLyXzx9wBx;ym{y`Y@5in@0)x)XV`77>P4z;#kTlw90J+NlZ z0&^Dg=NC0KsY8T0t|4GyW>${VGUCx?d619q3S`04`2YoTkkfYA+Srg|7zE{`?r-FN zO2vtLGEn)%ERz4`_0WgGmB^;^KT-TG~zjmMDHx zdE0`kKbdZ!G#XIJ$|F5=yXDu&%F3QuQt!WWm$CdeIu2FVvW`P%+XyE4VeibVwFvxAuyb2pVV+`G&v_m%Q#}u$YuUJ^Tg=_qlxfsrALO)>| z!n7!yv#O`=txFJ{LMquyTU=bUDx2!7I)C}sy%&`8l?cZvxbY7Ka?l@D4~H*Wb;Arc zIWbUoMaa1Ejh=uHo&FL5RtkU9$d`&gqf&Ij*WdpTO&{!`-w@}vZMs4m(0g>3+6!CM zKc6$7^_*9~Q{kt)b8M?Fk&3~r@9Ooc8Hv-(IqH_T8wW|<=hWdhti4FLKs({_c(Uea zp|kuim0f!@mFXHcovDU3X>^!!N#``5Mij=NgqboUeNwb%%U$Kt#l7UVk&c;;-1DJg zD{{Zft+3ln8P`%FL|aWLxomPPNz?hgtaH}+>sQu`~IHa^SsaVd&+%; zR)w~_gE6{O=Q*<@nb_8{{PD5LsVSvX_v}m7C1|bIeZ4cR`#O|q25yfa7s;pi8*VEK zE0JMjQ~7W_YXi}>W(6R#Zr%a#ZFLD|fWrcQfN#k(B<>0y zVL;{SWwUQ@-ke1Gxn}{kG$Y38_rYRBrm8t)0^xEVqx=ws?#wpXU|SPtc}ew97dGGB zz(a;7E&P@D*_d5j+MH5<`&1Q0_?k_s3nTr71s^ziy1FeduFA{H%j`Iv^VDx()fH{8 z-Dp^muI2hT74tO=I&Wrd97*i(llu-qFVwbeGygIo^S9L$C{`QD#4I+C3l5T#bbN~- z+UGnilTu-n0d&KUS=dXJ*2{O~gvM@qcFZ#@v4z`?>bYQ6M#jdEup6Up#zIKsK%()c zckkYjby#Y##mL0u(qQGeT~225C(TYQ!M4&eJ)E3r0p2$wZRyH)K~}&epW=70rxZ=J z_PPUfw6$|l&a2EwIB9@t*8AAQOzv z2zkg#A@`Ge_(90RLLj*(QBl#n1L=S7`*MZZF3bMB_J3}OHSjZJvE#))|2%%Cv0>v) ztFog}R4$&bQ*}4a;b)+rV+-ApQl9wef^TWrY3OKDQm&WnC(huKElo{ooYl`pAe+K~ z1fw#?ea63xgR>lphG{I<@c~rJ==c=0zVynJcBnW~w}Pkl4t~?uxg%zhpNN=BTMR@% zQR)OlcEvvXOzl4D&%$J2r8=W*9lWJ+y;-=GwJiIWsp$L`nI#@ld;z(sl1GG^YcX%2 zjk1kfzlj0lgzoq)8nADSh`UfP+zPZY~~Zx%QrT`$zJQjxB>mqDvCSNTr?Bwy3Snx0vhG~}L?mX@AJ zhQR5mL7Ln0lGr!MP0Zj_SK8&e#TH`TSXl3+PJ4O$uunvGP6Bqb5ukw(zY(BUD~t~o z0}D-qw_dEbkFBi@4trUCV|AtHbNwo}%RbpHIz>t@#tIVVpZv<#NBzZ4iB&Qh4-qcb zge^hRSb%eHd9KVh_W}r7)mHmWrV0A`W+8CgJHt{ggnrsigh4u>tiRfy!y!^WpgMVz z^S>jgEH^&y8pGr9AS0iN93E4AFfJt}1^wv=@QjVQdcS(O3rbf?KG8U=IS9@5HrUjJ z7s$(O8ij0~M05om0Mtj5?J?!%abwvZHz;W_2Uo4#w>b+b(8D^2bfz5f~BK znK^}>W5ghhP!Y=(Jb5=GDIswn_AUv`C@Cz?RE|cG9Y3$Kay#7)5i$iDooH&P&Fbnb z5R#nh6=duQ5=(=XK{5)GN2;rp)YMFwM>ip=yPWTjDrM`jIf1|i)&xIrKRndAt>Mwp z2k{#-AeE+DFB{^0(O8ueEpzZ^92UU4y$72l^pF4vM}Y?>Q1*i@gUQBq`% zVrMooRy%QR2wh=~J;Y$$-*Ypwvq@(LF>&!OoR4P6XgcR0dkHSeg{~r7<+7f@C)!0e zRcY`BXNNls0hm*-Uz_4$<-i}?+B5veC2)YS8(1h092_3rY+_;p8S=fv6oH0#1bAS$ z@lkQ-3Oz7zQyN=<1!kJ0b|mCZU0u`w>I2v|HpziOK@q}wE|;4R_}j|>1T@?Q+v^(} zpYVMEAoECpyz3pfe}A+!fCFfi+@ZoBvLvnPdd0gtFav(9u5x+BMc;J z8p2nZ$c4>?9o9VlaA#Xv+m67w>zEEn_<1OlXoBg1uV8L$Y#jS8+a$%E@9=3i3m{#-go3J{z`L0Y$Vox_5H2_;SO$G z^Voeko>?`Zn2-th!NsI(&FYq4NuRcu`6VObCgKF%-m4W(x|Ez;0j|{4(_@Kt6zX}_ zcX!lyE-wZnK>NpZqVQ@t1jcEG(@)&;-Dl4Z>G+JPHCQ?wJC=_&Fh)j3ggxB9s+W(O z-owCwJ1X!(-~tc#YTq-`-3&uynjKjnDzOs>a2eaGQwEUOZ%0W;G00qx%`^%DU$q%- zNSHCKpDL_vBVf2us^MsvC4m?#t0Z)xIzpU<@p(LLQ=dg(OuYj~OqwVXlhM!P`@m+n?^)MHl-FB zb5HJT!|T`9*6z_85(*0u0n=yg;)d|Pd0I^;>vgsEFZrZ`LS@vq2XRJ=7DbCC_PmU{ZBD?&aT=%TIPbXY-J z)WUGtvSsDLAI~UIlptZUt82v@YSSho25aQtm>!jj5r)Uam^(W=$9e%8JJ7wY+<*Eg ze9B>D--r)YY2J3R{Q-+C{KNW)jsnYXLx+qv#nlBM=%2c<1nB3gJ)ehQZ1xLPRPi)= z<>;>jwrh!61Jdk=UeCmI2p$iAT#QBO=c%_8cae8cDOFSj1G-16mD(#Y1{ zO+XJQt<{YZ^iddzgCN-nv9Z^XB=vaw7C6EMD%6cKO7mPp46D@7i}p|{zGC()o*y&= z2i3k&l5*AU$}>~Oswb8Thi<-pLSgSRM<-Nt0g*k;8+7u*y-%HD0FM3D+3O2{pl1(`Rj%|=L$8NAN_ zng8CRaZft}+6pwP40~&XYoTGZoQU&iHQP47(Gm1+x<+C(jyGq2I08q5&^zLprB9J^ zkc5>>Ht;gyr=v*lWHNwQ7mwc~&eQ`$)WnM+{997z1wKIBNoqLp*wrfKfe44bWLL~B zbRm=LMxn-5?$9Vsv)~d$4UqU61>rHNCt_)y=cWrq?Og2VQK$>Y!?W`5s2q#kv6(glRTu zD7nLZM6(VbPMy>dhxgzEhyb=PakdE$t@cqYv;@Ur~YHDhB&$^@G5$wKWWH3sH6=*){UaYVH-FQ+&J{NE_U+sCXJVR~_C&4Mx$)QqE%69n;5q=o(Gyi|e!7?R zLUTc<2@*)n2#M+`5Pfn79XU1pULQ_uJ<6M&o0~&}SQ?-l=lEVW2$;ixeH!%Bif3g{8j%4Fgka4cc{2_tdE+VoFwkXCfbV6serzv>0lDX z^M`Q&E4F&IIfgShGLi==ZDn0u8)%3dFl8_FYADo5)V{*DLB#VXz!nCaMI1h0ExJl0 z6r{um5-13+<4k=oY%&||X9#txs;ta`_nV~aUm7i^orb}K2R_PK*eZy*c^x^yiiF3Oyai;kF54U;sc5N`cYb;P=#+jfDxSyh!P{bo4j(;7O60K z`@c8V2PgJ!{Q1=Uk81 zkEaeqfTXl=WMh7m}~w>42(?^V%16{O+pQ9@*-JOV{QtSRwn%g2l^k&WtrF=)5@3<;yGKR~@fg zaN;oEn%tRQ>((pW2HqO6r>fG+Ou}5Y@`#zK_gWSUGj4`DHA;C-R)*)#O_$B()o0lC zOiP*x(&+c(JzVu1sx$^#1?D8vnmng!&iDwQVOuhM!>8M}*|Z00=JQ%JPekP?D=Qlf zwiWKOZF?vA<@uYItfm>ONj-j-w%azf#zAlMuCFp`FACHc-GdhlOPMw0c!cIBR~ya_ z2E9nc)p})5o*CL@QS*G*AynK_$797}v$Nx0KXpwx7Twt~BDVp{`u_Z_k6Topl!kfD zPDpm#GdlC>^_9T1u8OD1vZFOd<(UpWW>dPE_NK$#RYGCzgM~`$FE1)p2mHOJBM!~P zi`jL&m+TC&Yq@U{7JXtWQYAQFjHS0~!J?nNU%!4W$d;cOt|T{^o^MoGm1vl|ZqK=+ zd(A#RQ!UVpjEmcu@5{@L-&TDeu8>eip_RBuzjS01evWyao0TZtR+nr`9F^`_J5 zW^8V5?t&4Pp#XXFKmQ!eef2(6Mj~f!rax+|^Zor*y>+RvX-zl2Fq%AfIdp&Y_ICdI zYH{mdt5lzUCzfNQGelpqTp z2>mR_fl`_Nys@0IY^5n7VPW6PE951YeSWalbmis6EByTUWb$!y+!Bq>Mz~dl*_nQ? z!&9TZt-Xw<1o1XuAt4{G&4=;=H4W~MjmeGG?{OXe%*7|;coXYW&>8N&LLE-kMdI_?>)JBEzFu0HhD$r2e72BoI}>V?zuq)Bd#o|5X!ca1LFjN_gN&sB zSK-6`m+58lQ*jkWO3KOsSYe4Bf`Ur^T^}ADIH{wf)3xiE+=zNznp%|VnzS84LdvJ& zbpr3)xkI(iopG$l^_pYaG}smk2aX!@E?v9x2zj(Gn)saRcioq?IrFqkjYgw2ny=<6 z^cZWD<-=|6^>j+S=K!4)Fq0{a$lZW@QJ&m#yDT^{|Xm4b^FSbBQOT`Ol>lW4&#K0X=S# zSg89On;R5Ga;FD}1riYJ&t5bj69woyHHcZtm{-RwVSCI@beaaQ71p5^-xZF1uCJ}F zZCi8t+EdFpY)f_5K4sDDy_T9Sx+V4@imR zjJFJ@^EuW$$}iA}eDPv0S+%;({fqh3yN1#eltnVAatRL~{yNbXa7^$^E-hfIX%^o2 z#UT$vBrcbY)(1UAb<=k4*uMP;n>ku%F4pZ9UlpslO;|VWsK3`xiG;a>Y_J3;sk^&r zowZzbgicLBj1JFw&5e0!)}`US7p>qAY4pOX30XKz^)jN^L3*~d)!f{gH3^7tmri~u z+p|}C5#Vcr3|viJ=c1#+Vl-yn;G#Fq&W`0qjfrM{e!N{+IDTM2jAS|cDP(&;Wb8Oi zFP5I2O6DIr!Qz%Jh1u@R;roC&2_srxsLVmk$fR;S*Kp>i8^qU!7Z$#emaQG7S8a=g!EGK_tjYD}4-ZpJj4bke6j~Q&}_mt?LEeaIs zsb?WQx3>6*s`fY8mC1=MciLS2LuW?fD^jS!y=Gkug8da{B^zc7k9nT_^rvayh#Xeu zeEH*p_Frm}UbBQY&57|M6UWaV3AW1I(U8!+crn?c;@0yURUSTjw%v20*f5{GnRWKa zsnlLOlh^zgUe0SD71ht&i}lI&*4op`$h=a?v-kWj*Z3T@T)mKaT&5JJzN9!$c12ZA zY3e&@q#DJ&e|LLxYmJe=FWH-;4NgtcV`{gN%i)B(xYT5${uKSSDZ5-09u3BZ*#b6Fo;_m$1KN zbcU59C3$P5G&kZw8b;&{Zt4wXMU5uuf4p`9VWX!(YFeWC~$@?oTe;F2vZQcA=JhQjA$w;sLs z3yt2!r>UdEb4+$XF~qUXGzqUb*qZNa6l|Gx-N;`~+$v|H!#>ScV{)`NILxg&PFo{Q zAMagYC66UL&3E2{MR~D_IXZpgWGmd-c;GRWw;jpUz_S zdyZw<8(LZi>i6Qh*NQT~{{AybmZ~gI>znKMY+WmApcl~zSp2ZD%I^A))dR>*k*Rwa zb!M?=-`-f$UdNp3%Q6kNYRIg5)}%1~X+_sArv?Yj#vHQstWPt1;}gW3Q5n03(}zlB zth4hk;OZhh^qoIl>8a8*zALPIS5tWmxUr_W^PPTPyk??-m~@A=*Oad7(Ol0N@pqC5 zNTbYx%)WTsc;Phx&wC**9rwGMkVkY9!t37-%TIoec!F?hpOkL9bzsoX1s^cj{gb9+ z)^LS#tlo6uv2b})Hlxuswx}S>zN>=CQ`0k9mFO8bCYhMtazn7dvF-NZQ{(O*HLR93 zdQGr^^FR1 z6%2Fh%h)a`sMBAdug0x+VSoLFH_N!VY??1E&Zlh)UH$g$+X0gLfEQMCarw(no+rq4 zpw2we^48eFgn(o6QXQqCyj)yzuEU%r1}nRI)z+il1FZ#z74rZv*I6?M4jl~^7w(ra z0jRe@62K3TyIQU(OrH+yoO~lNd&_F9hn1+>Is**8jx3|MnCA8n!1!{MJd!ZZG%{TV zJ{(x_)<-mF9hdnzhr5A+ZwXG34viHlm2AD4j)g&PM%3B>V<4u4T5xBy0x@(PHBL<) zt(iRuKAmB1B06an>j_Bwf*9RpQW~;O&edKzcKTU_-;FYbIa%QdweZfPbgG|p_cP~~ z>)R@F`MjnFY~F3WMk-+olFE(itF}e15ZDZrff1zPyYY%JfeXL8;Cf*5GeN1eI*GoSQ}`l)HuZ9xe;_u6D*T7e&5r8voB5KynKV!>8v{Uuz)As>zQJKQ|<(*)NA}k&`hJ7 zYo0gPn$j;`C`kc8RUgN*)}RE)N59hbuhzd~9#5sUvdg5yh4JmYTuCFr!iHuz2z0!d)uh~7>3h1Mi zdafk+BYKlwKd6!EeWAvj9xU#f5lUq4&cWY?J#B1kqz}lRDLL9&Gz~C8V|dN<7_v-5 zQqGkqDNJ$e+n;zEUUFc$6m`)up4Dg!t|JKxlio%;G%*1osBUGkjAt(@(4n@Oc;z6G zUFb~D*Zekj=6*{{%Q%wCy~eWmO9tVc%+HT~Dtbw7iaixCxS?2_ZL!L!(ZwV_->|CF z<-mG`=uI9mL7<$(V_~uxc(SD7o8;#ex6smatTk>Ai}e-@FG1%Y*|YIde?Gc+gYEV)P_C-0Gc)*dZqd6rYZWb?%Zf_iydosuFR7?>GgvsfkSWhB4u^?r83 z*%dv=;Wt;8MVx+f^@Ntzx+_<%l%ru>VqGZhb2{Fiq!|Kzvift}GIp;@O>%toIMp^h z*!K~VLpi7lt@I10kA_I@BK`2EAz3^ULUf!#8|YISD#LOqhAeW2GiI%oJ#iUG$y>`tSQfde-st+Thv_8DNJjULOv@s!erL_Jzl(&Y5oeHFc9k{kUu&?R# zHNKWFDP^ktiP_Vh@vkl~j$&s#m8?!3&DOwO1EDyF=0BM7*b6E{Fjv8fIgo!3RM1U6 z8F8=`{Z(D=4rD2nxYO!gd*X*Y-(>)&2+eS9VK7x4mcFvQVR!fL-F&o#s~7Xx9|v9# zA*DmG;01DiBwI{7JwB#v=<3(JRmbuG2REH5ar=s`=!YEKnYv@gj_bI=-z-zo+*OTr zQq9)jl6}3s(eUIwp#_M)eg~?rI{YGoAGq9rFj@z}U>Uu;wQ&G>J%Ll#JlM$-9c>_9wSv%qq@hBAGhkzKWN}JGqTrP3RtE1ia_cJE;y21wJ?s0 zt0B`NR^w#~@`Ofa|3qhanT*rHJx;7Z-BqJ_@OHihNjbJ}*hL|>MLW7i)sD%f?k`ig z(EQSj-B#V`iHtEg>hPszmm{hzDjFUYs`a*Zzi>G>N9I_g8kos0d4GS;wuFq1 z2Tax&ke5F?BMr78qNK97==V>-);UrQ@<@r*M#JVz1a6mOXK84xWz+N1Z%9YtJC$g* zKdrtj&uhBC>*KRrY_CW*$f<#?OGnKI@7W3T{Poi+tT%co0$kfId*}bPrWTkkM$XG> zU7BNgLMmx-kzj3H6&UXebg-FK7>k$jsP|3?R-&g#Crs5;ncSEjCx!_BDL?gP6(Jv{ zdkp6UC6;=r#aNv#t{2#4t5Dn%Q|P~P#HG|B5FqQ%g-g$Xg(WR%wX020-OlG)yy2dc zrj1T7LG7J1{dE=_;0YT^*B6^Oy-v5!pO!Ia2?3xcM-NVd+-wR#!k7Hwr5 z)S4*dl-7gH8^EweIXcDw3e&bU1Ou61Uu~%(QWYOdpb-ehfW|p?)9hVM(HD_vpq^{2 zvU)3~%uV>KqfiGFvmAR{>g~#Wk15RT1+Wh7&4_uf>fj(2%Zd^Fsw*^h*dN?#5ZentF_Num=;2_{^%t6| zrw+#u9=vW}k`AYK#ukAv)lNvT$a_%x27BCqw8T6TRvV&19d($fVLc8BneRN}V^ZWV znR$ezpp2lRttM?qgj0WGQcptGq0^^NlXl+b!~Vr@(UAsgvNE@t?bc08%bA_*8TX$X z_a9#T{?UPJr=*PJ#$)2GiFpoihC_GfJtIFFgB6v~f;HqOWX58C=ZCMASv`-z zvvH`@?2ybSPpBlvnE4m$*9c1d=&z-wdb&5R0PyVm<-hi21UXy3(wJ!x)s>M-`#P&b`*$}&9McZN@s1Kjg(u3tq^%m&+c8*R!Da=#rP zXf2-k<3+2VOk}F5xtT{3caI>bZ!n}!Qq$c=zpQ8JJLv}u4*w9G~pU^{r6h2Jx0-tQ^Pdwpf8xm=^lQ&3TferNRw z;0>{_pmmMbBpLN=_){M+*9My|kDw=J#g9k;s>5Gu_ek4+J|6Y!jW=gl0h+xwnrG`m`Xbd)Baje}-i&t=}2vO;0{LaIV!n?Z(6 zfA-j^sf?JvxL;p1b3 z1k%i^(V|mbmP*9C{WJmzBZTo=8QE14yQ;4d5~FHW>9W`65pH9Brt32Qc=abKA@Nh= zW4UvoPS?jpJqLX3v+$4z6K0(=aZdK@%LNWPRKc-(1Us^G@FV>X<^wHncUjiG@YYfb zGiu4ZjwX2g)3qR>@)MdFPvy*Zzez_Y7kMl^XMbrnaMTlaPxZ5eqgLtlXjg2P{@ii( zfcb)iB6aorhr7*9)}1m2Chi)2bd6u9HOIUq!e2fntpsF4fR*N{oYut(3~CPI7wJVcwm3kv`AMp1XRR${wEElN%6 z3a{}4u7qasd8! zrfCPOW0Y-uj6oPG+oi&mBGp&#n)f1v9J-^hmI;~NF~goHqG!;Di^N2IT+hWtVrG(2zc8*^6UY)tdF+_+xzdpP zn$@R0+O9s7Xs}oF6$#Bj_~B>TJ?8CK1I!gf1EXp2-Ko`!PH1GHcfn6w8N9DI`^%oR z46bJ-CTpiJ;}ySYIMc1km9%xQ2ddd%QE*%XL%t8ln$UmtAT5jN+W;C%g$|rNd6Ld8 zX%)vB4Bp9*Eh7XXYQ4tYtIpv6CEh+S$^c0K(#;Ok&;({$u( zM^Mwpeio3I&GtgBugVnf#pXOy^ZazGbfh@;VAxo?jN_MeL<{VxZz7T~1Y@qGb9dX3 z6E#~Yw&*|ykeZ&10UypXXK;ZoLB1>ShUB5}T$zFt_Wl`1+j9}6ni~4o z8yN2hKyUwao*iL<5BsNz{tEs5n-_pTFij3`TiFX(A;}Mt%Uf%xaq`s)&*zOsp`iPw zyN-F?Alk_6aFl{Zt+-RZ71~IHLSNqz9nXOFMcrs_JOV8tO&5ExtZz)dAn>Qr#Od{4 zKed|`X+8z#Z1Zy7VnxL~`W9s~^>b%FK085J>(5{dGmmdUAvm|~big>^(E~+AS2r+h zf)-<6->le%H8srw&}LM~gAcfu)&r)11PsHAP$=l#D#XF5nGmW&X-XH8nk{~}!#Bli zZnhBZxnaA}6zE|3?aha`qSERkAsXYxu(u$vqzD|4s zy+lNztR=w{s}Kbtx^)*s#%zRT!PV5cXT^1;b0zz~$)jwO1{#MX_mu+Vg5xvh!%;L?gxrwE_SI*0nJf|?Ov+k37oEL0*5L83Wv z&)f|Cl}o2JD?xuZ2&W|ic0(N{m5xN|P3wo|w3X}j3$rT09U>yCn>J;%(UryDr7KK- zeZ#a-7<{*RO`0xLt_p^A(~Aax?7>m%cdXsCLBld`1LsFY1q>kdtG(MPzaS~`)VQgliqE|!a^AR zkj6`=tc_F}W7oK#Ru1NSRB32#A>;xp0resSKx(K?HB>rPwett$g_=R1k8Lu4l3*(Vr04MAmYKM8pwp zh=C#rwTM_Yj1a!2GbP|V2yGdo_g5s8%X~jRUXoEVf1ku#5_Xsq;Xr#xcN-swnXkoz zxB=iQhSGRTQ`9*gbanw@(DkJ}2zvpVcFCAV>UcSBjvEwUVsih3^T4)D;Lh??bCwp8 z+xrJ%Ld(H%oJU8=;cUT<5b1|BwS+tFEUEL=QY(F&yfA@B++rOh=ri{HjdJlPrd}-; zP?vOjv74w;84n7pq##N};IfI@t|Z=8np1z=b2via?aNCGT0u7Wl3F)!@rGXrV4D77 zqJWFn$%Ed~a+TMLm@m#ZGA9;+S-W*$`E>$j;@5lRuQ$mv4$@ysd@&7%7)b$w5S3eDvkH9`MHn6nbPUaKgKzgp1(^8f+RAnh%wil7yuv0sk zN*GtOf?ALR_t5WJ z-pF5I=JVks)JJDvjG>||*fX}N%8h(9o8pyscP9Up=-6u-6`Hm`zYg76ItY!H+_@@r1f2AlUS*!`?Me;O$AB-YPSYx2iQLciNa-G+_-$ zfXz)N&M9$Qy91E5K=V=#kAif45ENA9%5~_IydOV)?7KS++u|_xUZwVHlsrSimJ<;x zA7V(XaO&6Weh<-j-CXo0`Pczro&pAt6$Sw2MW98lBOVPZjW{lLA*C=UYd0VEdmlTq z49HJAy8o{U_7iLk!nIx5qi<)e@?b9w4ie1>d1WOfI&isqEfKPVz#z(`^7J76Yki6< z?zVzjVV2ri6#7b+t&4I&CwW5pOjmS54ve(WYx2NtVtUxR7vm-y@uY4 z3zi>&?yJ^Wum|iqa~62*RA)|%8hBtm&xv*o`7jjfZQ<^Nw8kcsSc8Pby3%oFx>!nU z@LtoCy>_mpQl09O#lNvois_K-G-{8{&CUd%zHK=y;9>tXMc&TDka%T|sY-V}5u9Qo z#9~AA4Rtg2I4eM%YP+MR8a+5ou&ufKEJ&p`J~p<-a9RkrM(C_1R<(VSeSD8|6NoVBnXxmy@zzZ_ z?WJPIMR$mzSH-W3`amMK%8&0G{hala*!d5{#<0QAbHbRkPc|TM^0B4(P|T9vESJe6 z93vE^KXz(8f0z$Q2MT%MquQ~z;AmG~UbO56B#wim3(X#>wt2s+fFtY>y^L_`sXXR{ z_yaB{9)Ub)WnrcFEWxT1o&@(00YFdoM|#Ht^Om>VhBpbR8IH286CS%87f_D&MG}zQ zsQKk3j>tnW&)4QxRi!^C#K))A30h%;5uWp=fPy@D&R6G9m|@?K(uY4-&&YJV;`B@RRDeSA*M+Kt&^DE7=ZYpM%!pCG80O z8#~|-)I_M9eRqpFMHrRPsTftPz!_*)1dVgD3Hsna;1AnQ^kxzR9LEt-bOE@BV5j=4 z4|<71>}iSvvP>K21CiApNLVAdhrM~N(l_Ffrlj{}_Qc3_p3=2kLyS`-yOR_^%zdlZ zuFV(8`m&N-K2W6$SpY%qa2-v+w^oz@UtIrNT+9Yxofl-kauhCTtT!MRzP0uo3KleB zCHfOv7R(HMB;yYEvL%T01&)p6ZAiF6L{qbAbm^=bpT-s}4DQN$)TB#jbogiXc9ew+ z5Tv~U9=>v9pzYhYd!t!Wge#%-v5&|N1aLmFFMN7@n~S$Ibvm*K1;!nYpIwe$YKVGI z_iSRIKKnsvUI)&~9exWNG@y!LI3GilH*E>Oh~yvOrhOD2dghio1(cUMR0&Tb;x2Zz(;X|wqvPFO%5tf1CchY zK5%w3QSmB!=f>t{4dwh$|0&+TUFLokfg%mnyR3clXBd2cK z^cZ<;e&bD;kvPsDb0rJH^$w61Z`uS3iwEdR*JC-ct^T;j`O?URo_Iz+d`mpW#>O^a zqB+kU5#>r;4FuOaxB5!7%ue~*}W4?vhz%ZZbkB%~m@ZQrq@aB2#aDHor# z{j;bO#2%Y*=RBoa8C+*88sHm0t=g)H%t>6VWb+X+fw%}urF-I>0Uul71m*5K_W^w* zFWHz_CCN{Jt!jB1WN;&VZn|?sZYSQdd5-lfjs7HOF?oN=zTW$<{tNzF&U^e57Tzji zw^U&O98U7jf61sI_y7ty7Z+*awK8mt<@|r7{4w;AEXQ`OJ#*%Wk`Y;+i`I$pjK)mW z0+@!!4+&x`Vb6MdYbBjID2FBgV~9u~%Xn`rrX(v}Cy(Htv=RTaxACtrGyj*bLM*O- z-pk2#()Q;)R?MTkONYm)YU~%3u+sO<{|fi@fBAy{U-5wdvsXc4@cFKIy^P(ZvP17b zS4JJ%;5q&Y&QYjXXw5Wg%ZTY>8LwwVxF89oUoYn*MH@qS(u16zgwQ=ge^TK{YIht; z-zvYK$&3DYGk$0hlaK?*W3n7Rmt6{6ziCqfO2y}zq}F_X9`eGMi-vrcI0sO#gAgDl z4npyqfJ|K+B$_yQ4=ynD(*Od@9an!OFMs~ZJz}l48!RU+-LZ)Pg&f8n`IJ#K3i%n> zo!IHh0i;#-Dv@8fxU7{|>@owsJ2aO9hd0q@;8P z%}P0Zw)Z5g8}m2pCYP}5?Ke==rLcszd7)D3l&+*em1uH&Q;v)_e6&s{Zgy07(1UP*PL( zMFv>Q@@%&;TQ5;0hVYp$CcEG%A&{OZMQhT0ytR_g(g1yOD(phgPGhy1ukwGUyu8{z zS4Eli?^nDynPgTKBP}rHD_H*g5eP**wpmCun$wJBY$C?G zdQ^nKK5O*4@NAsF?7s|4|7I?BUXvqHJai&oIc?fEqXqUN)PLsGVkV8Q2ueZ6%}L)p zyatuyRNY}zUVD&76*F)pdjoNg@NTD2czx$I7v%KPsWgViw)tnv7@3L(!sR_$U~u;a zT^RFvhAKM7rhTf^|iZAxfjCgZmt+<7^EJRr2GXwuSt{ z#RXJ#m?&*CQ&3Y|JA$qK;7Hm|>LH;yJeF~B-tS~nHxCn(hdyi`dk^{YtRXOp)H9Hl z0))&0N>wDjb|`g;{iit-AL^fV`9R-qfS!rk=9WiVYiHPBi0O)W97*T5!T!Vk6UV>C zhqZC`j75G<`MiFjqoFaI)3wYZhF^FiGk-cnNa<8cMjbaxspWb$gm zq1NcNa~`FJIqoHPajsQ{>!Igf5M`wWOPnQ=(Z~UIQ79Uu7X(CjXh6nAc*ONqtr%DR z;T?Y}tT_1dPvE(K$I1VcIQiGk+5MkJ>;K=cLjBDo|I7RrU@AFih4@pzFKX6V0Wooy z6*7qj6h$f;yfmAMHJP~0R3y)kopbTQRQ@6+X(7pIADkbXXmlkuW|8vnnlqrFh)?=3 z+Bql(>v!S*Bw_U=>eTNBl#6rz*={QmvJ!?AW|aaWHC-d_oK)(%u>MuCE!TFfN9uyy z8F!@rh~1>5@kFRO**p1RR%yDVRcvX6{)u&5k8G3nE!Ms#;_&kK%NHMhSZKd*yZSF_ zw<;<=m_G<<68KqrTm65ax|6RSGv*a zWQL0jRk~mlsS;?Ap+5(+)H#e85<_d_N*i8YUONEUtu!K~F|rq|BftGn$4v0+Y@btZ zafp=N$!K+3P#~Sa6<+KSG8Ra00|DODQUjkaXZ9&0Lkhre``K@^qJ7VzU$XI0@8N9% zZ)FLBsoP#SedMn&Uyb;@j*FJ99Iit4^&Y0wJRvx$*8P60fbz&3oK-uOdtJ zGOIkczu9A3u!OZLozYOtfFL&U?y!7C`@q$!S1o`MEz8>-FwYVNs}G~NwWOp3a-z!E;zg8) zzcj|q%O>KF^z718c%Sx4bH@22^dzH#trNMorsmu*a2P0l}zusM_flKCLYfx2U-V1x^$$=mU_@XAC&h3FWJwt3xxN zf~UJ9#O}RfV#e4seE!g{q(CpK(jk?^8l||9)wme;gys3Jj(4`*aCQrELKCH<9bT&U zFsuv_y&t+_F_e8W+-U|{0e?P38wfZb_SD+06Ybu%!GYC|hO*B&E<0O}@Lag~`$R>9wfP<7lN1>6HXY`q{by2F7B-Uk zQSu`bg{fLX`t!jvRiID`QmZl9advt_YUHDuTD1BN9?pQ-!;aM}opp6|cAHi_d6Aa3 zl|~|rn%I=>2IXxVZMLs}Rs_w1{&MTL16@ILQw`TG8$& z(|3gK-lN;3q_3|}Zv*pM2GV&5HWx9PErh<+-@R2I9KhPo|JYAIi_*vE)~^bSwBb}8 zOS+o4TT00+EETa(w)@=sdw;L{vVP`SHW^ic zBS`97Qg$6@VW1N=!<-SRo9bH4Ds*t|au|d53@X5&cV&|PH#pj_x>0wpg|H206aB%~ z5c_sM!1LbB)Po0%jEr=2bs4OkD6~(L^%FVacRaTK8>xBxWJBg<5=*Am*8RgZD9E;z zDiT|0Wcbs)*W9oegl;%Vi`d96u;1E2#WpUaV~mQZpJ0aiOTRtyY$g4t7v;FB6ZC_T z|IUFTfT>%x6zY_S!tq|s+(`{A!|h887IzRM@Yhdo7-NzeRrU3Hy;A3@rxhNOu z!Qfg_4^y7Sz^eZA(WjBA-09^@kIO;atB)L7Qv3K*l){WD8GH^3e%b>>iHx6v@Q;s5;$F#5NRMx1Id7>wNBI#Lc%n2*uP{Nur! zoQHEfV98HRl0N8d@QQiJk7bl$4&!O11E$8_LlV2EFw=F6Mi+JewDj-a;JK0hnaIY( ztNY#`Sz8Z|{|sn3wkQ{!FwT@>4j~y34C%ieGe}*Muu;u4?LjgB{_3a`u(I4D4r3xR z(a(Y@)_!rmd~Hv6Pmf)R7^`hh_{7l@5hT1QAD{f5hSHUpnW<-Aeoz~<@ocNY>}G(3 zGcdy5p{|#_@JFO-=mRo*mdG5ukTBx!9WiA~p4Ce7ta=Qac}Ow1?F%caM{AA5PERxhC z22r+N>xSJg0CDOv@>^}7#;3F;`FP@V_%$mlcW;_B7T{*eDR*uCGC0VN|DbHPj}Qq$CAwP)ZicbuVe-tU3hk!CW| zg#9Zs%LI6Y#(p-o)T4Dl9r-!M+3ku&Ip=Tt2FMyAuDpvmpZ9euVjooP8XIlS&l}t1 znjXL6&*JOsJw~u*rtKnh0*;Z?tv_nKL!G^f8?kDoQsI18c-;iR5kTZs5 zWdenOZCbBFC)+s}sepi9l-dza2*0e}LH#PC_54@bZkWV>UiVwGB}-gf9Bx1<8Xc1m zso$3!;)$fQ7$!A51l`l0{O={1aWdMX(xtbzm+iti_T^%o#*yX*ueq53D!AO{Q}JG( zA6=tUiL%S}AGi4gw@C)1s1x}TIk$0|N=|Pb>~`&FD4;-M-r?_@K(Ywj+=!L{?B-o$ zPHG6gK&MPA*oG$Zw#naysW7#pxE|9SrL=O#QWgw2C5Cjcc zglcy$P)IW}I(i3uF5QwQU?SH;N&W8Xtm*EgVglahjK3A#N+aW7#0L_ZfJCc|(*rtJ zi*Z^8RM)dUS0zZF0f;l6IB<>&Lzan$l#8=6X%z;f!7S0K& zp1733M=LzGBOUMjyVM?O*3b3yAcTLMewH}RHjnGDMN$W}$#;<5=uH00+f+mDrKP2P zpKYOF8kal}BZ)E$Js!qO+O^gs=c(@y)f7BGR9fWG`^g$Tcxer_5JNn=f4|`Q3T;=^`29` zx^ZmGeR0q_1c2>-KSF|*V8|Ci&Q(%VQ+x2>fh{DNNOZ?QGR64CAawE4WV{E~ z9UXk|<8eN@$>ojM_~%faGvIB6v(TR$r$WG105uo0ibDXmq9osx8>yFIY$a9>Dc(9F zHI?&L8c$Dljo0A=Csy+~<8Tc)fuyzw3k!$x=b{Bj!Bhzn^JW+uQ&^l6s9t7$h`nsYr_cm-;9Y5Sg0Y~`6cYD`aP%MhuTi2{~HUW;sWTf%LV7nBy^$^y_G1H2WI2ght%@>-Yj+F2ji|^8CKiDmf zMV3Z_5>t%N_vhtiar8<;M3QD%auQm{43RC6xiYiZ^hXG~vDkaNmtbhI@cSZy^Rgo(Ri z#P~$)=UZtQF%f9P`2oF>;_xRC2OQ0yDclL$x?%md;L+?l)trs2@5+0m zG3#-jQPsdmP|omEVZVM{h!lp;fr@6D<{Co3_4}@YRtp=^q;I1Q-Y;|CI)jX5UqVis zpi1GKuH7)&sNKc8>Lipk` z_k3%Q6D#6hW}&H|77|8FZk9B?05dc?oL~+vp-NBb)ab;-|Lpe|^ZPNiK&L`;HH#`} zN`E6klZUg_(IphcWH&7t8`a6=TLm(Bs(We7cg0~{45CIABw~n>T6kD~@b++jQ*Ln& z>Beb8|w zfK4@!I)Tg<2JV^DA336G>Djzp^R3|p!Bf*xkDY1^h9m=sC4fqLJMyBxFMhB6*nDzI z6FEW!n(<~b+(phx@iQU?k^@#sYr&MIpt1B*Kv8PUz=V31%r_W?ry%1T@C1HKuz;rt z+;Yd@X(>=bAS`V1>-R@C7Cha8nOn;9r zMtlO9qpCSLSY<1XIOV>eWnvp~)Z>c=CRsl;I`yDEIwKagmi+F+NP9OQ4#J|u9QfC9 zZ~q_Ad;k6j@;op8&7{UWs#~D0i%}JFSPMhn4raZ`M~_%uv7DFjN3t}F0n0_OWvciW zknc6?_l#T`14D6yu7`7TCO%w~S8q2ZKcS=G zCyvGIKNRjEQ;!K4${^aW;$9fHMpygmxk)Luw=g+?$fKrmzyZqQPmp=n+IRTcJ@yVMC7x((ZDpXf6RP*xo0l;Oy9)Q##=fp+Bj)6)~Tg^F&1s)7B0;x*1-w~O+ zItUs^g$_|lzWH(Nac`_zVtoQL;959giF?Ni@X%H7J8J;E`AEDLY)7nbL@(8r|7_Hsb zG@dUJBL_JN@el>YLxm1I*jcc|C74lvGJbabj0aD{%$< z?eMvFz*!qVx{kaJrT+$7jdKzIN?GCGd50|Zf99Qkp@Q>2VX4iYjpBCq4Ulk#;6=Gp zL*|s_=IdWr{OuH9+D{yHkFsUUH&Xu2oqr(c`Tu<{P(P5m_?ZKGGD7U6B27J9iUEpv zBt42;*x1pxI^=}6aG|ejvD?AMsJp?9mf4)mLwPAP6c(-3FfqG?xllLD(GpxerP&gs zNF3e!eU>JZk>EDd ziRm*sHa2$DnS=%1a%sYZ4FpTj4Gv&Kc7sK9h80z68-*6Qe zMRMwG;X-b`TbzQN3_wAknKlMUmV;pSE8tW&SGqy=V$w(kW0@=Nk;Eh@!G6c8+YON6saF(ld8?InINifxk+$ckv|?Q-FJ42FJF!W6=6X> zv=0C%TEL;5RhHjFoHt11!^dG_;qyAd0IdrhLBnO!!>DfI{wk*R{LG4)ni`u{HES1@ z9~$KnSG?^m?T^bA|5{yDWsC_fZA=A2u&J!6Nq_s`I|~VnB)7?dZ2J7xR#vJ2;*8FP z%?;~uY{B7Xct{Sw?s2^RI)&mIi5qiZS`SVM?VA{`6Xv0x#ZWD6Ra z!LX8V2H;dYKsoSEbSj}Pq6!2s&Q1MRSw8F;>3->k8UHP`wIX`yZc7(86uAYx!wjv)#1I_@AGIwmZ#)=^=qh}&mDd+xEQv-x%biZ zg&&?Rhj-+U9(ZWv-lQ#RMCgAjtM~uOx+^p6{TcsQz;g0$1W^84i*(iT-g%UVKaxy> z-4ZLe2n`VfGYNi)LV-%4M`I+HQTk8m!+E0)lWR>Qjvg^=BgcY4#l&fRA&|zSi&o8} z3?IdTx$g)e&*A;aq`Kg-A6{mi{?_3C6Ausz%A1D^qgv=lU9JLnHpL@>adE+f9 zANI`PNWDVXFX&W(@t2D$qWs-nQHm7*0R+5+IRx18lm9o6XWt_P={;mLW1I}!Wg6Zy z0WQrLj;#!!_!v1Q#z=Rf1W z#C`F1$lf-D!{m^loHJqyTztS`LA8NcujzKjs;98Ygz{`lXtq}k$VrYc4?V9kl`XV# zv@LwBQFETU|I4NOK9Lz1asuiExTO{m@A(t*3-%3=d`pgryT995slW)Tw%pQS&&FH{ z_wY-*O$jFam;*5Z+r24zbe&k)*m{atc=%ZVvL=&$o|vBi-QXVJ{m;z%SJxiK6j=a6OLlt=nCo8BTW7!+XonPrZxOx|ZjMg4ZXhUpp%cFqIVBwuWQaS+!eVl= z0-0^0(Fu`(A&g?UOokv&O$)P#Wsk4`ghdD6KnPJg9GP)$dVGLA5^iVSq?yg#AHA|a zmeRO1?L6g~jrEsnqvR10Zj4NSgQb%JL{@>6L_Sa!#|Y?#a311P!|~s(Fw6Xr#p;B! z?;YXT$-iAOQKXqe$9%qZ&;(gH(8lNRjA?Wtf`ZaiBjz6>lTG0KvFa{`vMNo(P|N_G zL$WtLf8M#+OL~l)Y=G-`FIXH(0?-B4P+KfFo<(UPdg`nY6Wry*Br7)ziPn82*07-^ z=oa`$X&eBYKWEXtn=6#)@haP#H^IF|e)q^!Lrr5k+L?fM;Kx5eTeDfwEtt+}%! zqA+vnVW0=c@7>uW@x+(N5$D>;>*S9@eh@X@UkQ&X{+>k>Em(v(XMcB{sOy+)KOU*Hb4LyK4N ze*_2p7z)AK8YlO=4pl=2YP7gmO7p=-e{xd{R$}nCyCI8#O|}^)KGROx+rI#vAMNe^ zV@bF|Q~D_KF6IdyDBaoq2o|4W7EAMF$9xy4yEss*0?Nvo_3L+R+ZN?sLZ&a3R8*=! zQGZOgF=`!$k6{fzf5N+WM(=84mSF^Q$vLZ&=MEM?>S$|gd-CDKaWRXU50VOA zv+fvTKL{YjB1Dg`sd%uMX$5u5mMx}`WyG^bs9h-3JTi`PyLazi&BGIeBaIr?VVbMq zfs|b&q~OPxaA?hF+mdev51LtJq%ZVUqv_^49J~mnFbd55ns-BsDb2>1e_KVXfHP|t zBTYO7I2`nhyS)4O^Q%9vKQldnvobSXM`ENMy4$`s)q}W>!EbigrbIEbv&EohSg@uM zi`~`n!H`nbhc0JV;XJ_mVF5!`PPaimt=jw@LTLuRQE6A~;7OmPl*OuJdbzop% zfIQMnr~a_s2MEONa&r0*nD=@f*D6@PeEF$;`!1G-$;HCt*d~-;R9UG8MJHn*=21XE z0M75zpO~1ybR!M#QhNsUg$EfVE5g4i0s917JvvO@Lx`J+t?ll~ zejP6@Eumkj(O3b;0>UYhewy;g*#hm){WWmm=h6KaYx%|neG$G zHhr)`(w_i@wD;s(x^&5rbsFMd5LM27{5Qi^-F9Prja%k|*B#bSRxi@oOL?i>QdpuG zSKrXEWoA;k5sPS?q}2v7=|byGL5-LrbUcdSewStZvRVqW;}O+~hKYE)Sa?$)s|4V9 zRO8{ylma$ehP}(@?HTpDJ0kkhEISM-ULSbVjlgzH+!b7+q^f$CiuEY#ixhw=7<8wW zS0r(lw$$Vrl5Qb*V`wvZgnFFh@$k1>%?mFE{S|SxH=v4I ztz;l7DtgUG+w10P#Z+3LpvExpvnh&8U7BUEhyn8%bjO z>!)ofwfFwfvX2LPVXQQ3$V?T{&wPS$)KFQMijGtTns}mWADbasFwobO8{T;@ZbR>9 z3y;}Raelyczp4a%-IM2z=_x2EG^E=c=<4b+t9ucByIIch%RL-Q@mR*G?=w#2?Z%4a z^!J?SJN02OIYIE9SYTe}63UVVKVghwS-q_$)i2kC>EFMi4w*0+aDJp4{iQm ztzCIM)cYD%o$4IwbQ0y1rNaQP;zAWFYZ2*DEY$U@6AyK$i(>EU+HW@gql~?Ik>NB5p17HN48Ygy6PENRtw({kQ zlnWQWBf@PY#H<$Q-l-K_H*~aHEOhY?X{6ECNlQzg^!9E-xQR+gU`}3#)`&j40ekBU zOYXyvv9!3jch`xb7FfZb+-KCdwn`g^JeH-MB%HD~YNJ&9{tOK4>F$=5k&#)vWC>iF z4XA5@aXiSmhK2@t*!$Ec3A`sxUIX>=536p!_*Qcq>EmIt5A`*y=~UlyDb(ayLKq?{ zC@6^gOj88D`BZPM_y+_c_&IM}`qD24_Vs!h%je!%7*M8zkaMWKriOssu>IQi_Yf$6 z=1^4J-|ldjMCo)Ss!mIfV&QxD9hUlbz&nQ7`KW~ZY`Miyy9cxk42_I-_^!}C25L#( z`{&-LVOk#67)KIAIR^1MZw5;3GZ1Fpx@ld#Dv^b-)AFI_{^9Z^L`6=@_d)?F|15sA z6Q2G@X^r%N03p6D2pxrh#M?_0<2ImH!6j3ypiotD+2WErwV&o6_ZDHS6CKqidlo$W zul`?oC)ylykD`bKx{?t*9V3MQqh721{99ULW2tVoY%5b{jbw zii~yGY!{%yvhqY;OE$cdAsgyF{lirUv3rCVxRszotGaH*AMiT{Yyd4TgkNhElxZ|` zbTc#|f#F7t8x>`YLkL6w1g!S{Q11SGsOw;})(Ao(cU$A|a(c3LFA9Gi9)5|X7#$Pi z%z2YVPYsEWkI(o1a1ikUSfTH@k?TwVNbTx3myW&2Olr3JV)q|N0LHjJ&M=yK3@L!? zG{z-tDr_gs4B)L9k~$u6$9v&ir_tN%?wirui2A!^#Hs;Oa*TW!2(8eqdf{{aE_Q8y zJbk{f3gHFv%Z*nZ{4(&x8sLYHj*ee_$I>C<){vqv0%)#OMeFdJamR`>xn-JYZEG9f z;lp1}Tq|i&FiptDUWX!uqGy$9841I{ns(t$sCf_VkKW?YA+d1_&00%K>tL?qMQ^-& zcugfOEqf@QSfM0E1r?kwA2mU^wq+f_TF*c7bTv`R+(Epqu5Lncz-g>+#2LiCbE9E+RG}=0&%8uPZOC&mCF+5# z#`|B%L8JHpHpiNwD0j`B$9rNxhlrH;%9Y8;q)`sVTP8$!NG5H89{Z^UHVG>URHk1mZAcxkk88a zO&7<%A}4-V046aF$~lhmbAp_4pb6Qd0LRN-;sI9|h|eeZQRre6InQ{>e+&6;E)Rhu zo0tV9NSY2?>R45+(0ikq!lwQ!^9p*NhZ^ zoE_K@!zONMo-K@UA&UsaH0SQzjC&4Sw`?x-r=C_-?EyoGq3G-BiM`LqcLHRC=bltZ zC@HOq!Z$V-qje0}pt)O!kkHVEn|rPZyEE}k9do_;o2^8mfkL1{RCw94C<@S%G2Pjt z0|25=A`$b`I5iYb4VMfJU=z4iE2zozeEfI^U`BvpBIdffWvpORV`G)>Pi;jvky%WK z;vdznI3FM&2hssrHqVeAL|fmc0{2#HNM4zPBCWA1i1OdRC-N{tPAWZrhR`FK;HoH;ODWi z{Ih=GKq$+;TCVQn>q~_&B7AU!(@ZGoxe?EgV&0QC0}x(ZN@_E5u#$bSo&es>`S(@^ z|1Tzk*E$U{;r=}A=k2@c&IN?LHe3@snIpE2s&c(?*0Dmlp|*00E7+Ol63Akn|nkUk_Pr1+G_-;aNl; z)3)g3!3Unnc*ChUX5F>zdr}Rf>mjE>5&{eiaK0=9H6mPXIu^$usvgLT2LzGx_xC62 z5o95?^?vwgywmmH=1g8Eax1ReI8m#e8O_|i`w|rl0tIgU=npAD72ToJq1lb8Y?AIl z&ghNpD9m*6(xn0%nFIh>n_+`Zvp`wzyzo~9DXT~4zPOF@yaX||kVpv)U50q)ibb+J z-pbH01R*==&71w|zx^hF1yJ|hiKh&}lU8by?Ph4 zVKYKnBhu1R>L>MZzxRsXV>iDEa!m>Y%^|%T5?^(Kez}QP!p5MoL_c?qmx(4iAjZ}k zGBDQ9SAk$U`#H(aUv128WMeI(npow2B1I_hx6mn?Pm?@Q8yu{ zwEe^lz&Dn%s%leuE!o>w*mH-{n0>$k2`e*Tt|K|hT)-pm);$j%JU9w#s&OY+DUva(<>cGU)xC;-fM7EWqS0;nT}NRgj=8Q#L&Rj=grN{Q?pLvTXC7fCc$M1 z_+4l#qjDKPuVDFO&OPQr)ia-+pw(N-iVcW%q8yU?;rs7jg3=YhlKVtG+}#hE8X1LB zGGPAd9lx_lj@sXU6!2+NYthW)Cs!0+li31C2Ez*&Hi)lr?Onn1WH?oIt*UrD=Bh4r zPEF5n2TrTFbKae&Gdp-0ScdETpVtXy*KDzG01Y^4R2HLTp-@GGKcgI(C_gVfa$u&x z8N4rw&6XuH&60J)C#!vB@>({5(oq-?!pVlud+MgzYU_Y^1>tCOKx_*~#}p{eyFTQj zC(VK_Sr}OzmQC2=>|#I8haAhMrbU1<$W=s@KS_@&8rhF%9~~K)j8w>GhevH)1*Wbd z5%KYxh)m0jdRFjhOL0L@4@FH)EtSjdN4iP9Lr!oWDWx!)DNGu2yoPMBr5x1%q(cGj z^TdR8uxP3B2=`xA+4h{}jTCcnO;j~|aw#yTw zJ%t*V4XGom6F#Zs`7VHxfynYgUW=4`j-dp7kQ!>QS}R8&X~&@ejSONT3}RL*4Cgqf zE!wTLoZ~KvNvX1gw~e-w7kDcCxdC=`BF)cW13n<-&$&n`T!);L*GRWMQIGn|lj7R} zqxqwd&6Tu=Y}mpe(mu#IOw=PqJ8s*}a%_)xL#DXc@2}!=0h_}qI6KKC=;*Nz<3pT2 zk^;rL-Z#Rst`|o+KiIoEPU9JYbe!55ggVCo4C4$5508n9voUcXw<9kG{QDx%j1DW1 znX2>f<7CAmXSDXO6A-K&@?4YyZgdnUYzf!)o81p7wrC3IaqZ#7Og}yg(?_qtW2V| zzdZ&|vbiX}mEG3~%}=2%tLLFjNJE3A+~eP%=i&BYvE&K(N2Eik1aeP?;hUEP1_ti% zoAtTztCu0UYLQ5o^KcqIX6$@aeoCJG`d+Og5+$(5z+KtfvV7n)$Ry(cDwB&ny8=t3auE>=AFlr`)z}bh7PYDlA%uEYAvtD_%oEizeSOWu&u}Ql7#~~M{s4LLn$AvuzsZkwSk~{SG)erTW=W#d zB(n>bE*Y}h+S(8;CyjE!TfiMbU196RXGWb1U6E+bgnL!g(69x9h(bJoishZ*VNbYD zm%YzdpE}3H^g{oakd#CqO#_FhuB$UfCO{ZjJw3e(SkQU$r{!xZD)fOM8-Y$5$>4xA zFiChictu#ww}A2#aqBpSD%C%4f?t;jx&%?D>t*HTjcLZ~5c#k~4dJ^F7M)B5k_zil z#&_pm96pUKbg(EpDB mkjMM4bNxLN`Fn0L{!Pel=iR0ayV~cf76^S4y*s*x0{;bTtU8ha diff --git a/docs/articles/09-convenience-features.html b/docs/articles/09-convenience-features.html index b88f23f9..81ab3af8 100644 --- a/docs/articles/09-convenience-features.html +++ b/docs/articles/09-convenience-features.html @@ -19,6 +19,8 @@ + +
                      +

                      Convenience features

                      -

                      ZJ

                      +

                      ZJ

                      - Source: vignettes/09-convenience-features.Rmd + Source: vignettes/09-convenience-features.Rmd
                      -
                      -

                      -Convenience Features

                      -
                      -

                      -Graphical User Interface (GUI) for setting options

                      -

                      I wanted to make disk.frame as easy to use as possible. I often forget what options are available to me. So I’ve made a graphical user interface (GUI) which can invoked with

                      +
                      +

                      Convenience Features +

                      +
                      +

                      Graphical User Interface (GUI) for setting options +

                      +

                      I wanted to make disk.frame as easy to use as possible. I often forget what options are available to me. So I’ve made a graphical user interface (GUI) which can invoked with

                       setup_disk.frame(gui = TRUE)
                      -

                      The above opens up a Shiny app where the user can adjust the disk.frame options in a .

                      +

                      The above opens up a Shiny app where the user can adjust the disk.frame options in a .

                      -
                      -

                      -RStudio column name completion

                      +
                      +

                      RStudio column name completion +

                      library(disk.frame)
 mtcars.df = as.disk.frame(mtcars)
 
@@ -147,9 +143,9 @@ 
                      
   filter(<press tab>)

                      you can press tab in RStudio and it will show all columns available.

                      -
                      -

                      -Insert ceremony/boilerplate into code in RStudio

                      +
                      +

                      Insert ceremony/boilerplate into code in RStudio +

                      The below will insert the recommended ceremony code into your editor

                       disk.frame::insert_ceremony()
                      @@ -158,7 +154,7 @@

                      # this willl set disk.frame with multiple workers setup_disk.frame() # this will allow unlimited amount of data to be passed from worker to worker -options(future.globals.maxSize = Inf)

                      +options(future.globals.maxSize = Inf)
                      @@ -174,11 +170,13 @@

                      -

                      Developed by Dai ZJ.

                      +

                      +

                      Developed by Dai ZJ.

                      -

                      Site built with pkgdown 1.6.1.

                      +

                      +

                      Site built with pkgdown 2.0.2.

                      @@ -187,5 +185,7 @@

                      + + diff --git a/docs/articles/10-group-by.html b/docs/articles/10-group-by.html index 19d19535..c2799355 100644 --- a/docs/articles/10-group-by.html +++ b/docs/articles/10-group-by.html @@ -19,6 +19,8 @@ + +
                      +

                      Group-by

                      -

                      ZJ

                      +

                      ZJ

                      - Source: vignettes/10-group-by.Rmd + Source: vignettes/10-group-by.Rmd
                      -
                      -

                      -Group-by in {disk.frame} -

                      -

                      The group-by framework of {disk.frame} has been overhauled in v0.3.0. It is now able to perform some group-by-summarize operations in one stage. In this chapter we will cover

                      +
                      +

                      Group-by in {disk.frame} +

                      +

                      The group-by framework of {disk.frame} has been overhauled in v0.3.0. It is now able to perform some group-by-summarize operations in one stage. In this chapter we will cover

                      1. How to use one-stage group-by
                      2. Manual two-stage group and hard group-by
                      -

                      In the custom one-stage group-by chapter, we will cover

                      +

                      In the custom one-stage group-by chapter, we will cover

                        -
                      • The architecture of disk.frame and its implications for group-by
                        • +
                      • The architecture of disk.frame and its implications for group-by
                      • How to define custom one-stage group-by functions and its limitations
                      -
                      -

                      -One-stage Group-by

                      -

                      A one-stage group-by is the same as group-by for data.frames. This would be unremarkable, if not for the limitations imposed by the disk-based nature of disk.frame. Before v0.3.0 of disk.frame, one-stage group-by was not possible, and the user had to rely on two-stage group-by even for simple operations like mean.

                      +
                      +

                      One-stage Group-by +

                      +

                      A one-stage group-by is the same as group-by for data.frames. This would be unremarkable, if not for the limitations imposed by the disk-based nature of disk.frame. Before v0.3.0 of disk.frame, one-stage group-by was not possible, and the user had to rely on two-stage group-by even for simple operations like mean.

                      However, now that one-stage group-by is possible, there are still limitations and not all functions are supported out-of-the-box. Hence, in the next chapter we will describe how to define custom one-stage group-by functions.

                      An example of one-stage group-by:

                      -result_from_disk.frame = iris %>% 
-  as.disk.frame %>% 
-  group_by(Species) %>% 
-  summarize(
-    mean(Petal.Length), 
-    sumx = sum(Petal.Length/Sepal.Width), 
-    sd(Sepal.Width/ Petal.Length), 
-    var(Sepal.Width/ Sepal.Width), 
-    l = length(Sepal.Width/ Sepal.Width + 2),
-    max(Sepal.Width), 
-    min(Sepal.Width), 
-    median(Sepal.Width)
-    ) %>% 
+result_from_disk.frame = iris %>% 
+  as.disk.frame %>% 
+  group_by(Species) %>% 
+  summarize(
+    mean(Petal.Length), 
+    sumx = sum(Petal.Length/Sepal.Width), 
+    sd(Sepal.Width/ Petal.Length), 
+    var(Sepal.Width/ Sepal.Width), 
+    l = length(Sepal.Width/ Sepal.Width + 2),
+    max(Sepal.Width), 
+    min(Sepal.Width), 
+    median(Sepal.Width)
+    ) %>% 
   collect
                      -

                      It is important to note that not all functions that can run in dplyr::summarize would work automatically. This is because of how disk.frame works. Please see the next chapter on defining custom one-stage group-by functions to learn more.

                      -
                      -

                      -List of supported group-by functions

                      -

                      If a function you need/like is missing, please make a feature request here. It is a limitation that function that depend on the order a column can only obtained using estimated methods.

                      +

                      It is important to note that not all functions that can run in dplyr::summarize would work automatically. This is because of how disk.frame works. Please see the next chapter on defining custom one-stage group-by functions to learn more.

                      +
                      +

                      List of supported group-by functions +

                      +

                      If a function you need/like is missing, please make a feature request here. It is a limitation that function that depend on the order a column can only obtained using estimated methods.

                    @@ -217,7 +212,7 @@

                    @@ -248,128 +243,128 @@

                    Functionvar Exact -var(x) only cor, cov support planned +var(x) only cor, cov support planned
                • -
                -

                -Notes on One-Stage group-by

                -

                The results should be exactly the same as if applying the same group-by operations on a data.frame. If not then please report a bug.

                +
                +

                Notes on One-Stage group-by +

                +

                The results should be exactly the same as if applying the same group-by operations on a data.frame. If not then please report a bug.

                -
                -

                -Group-by notes

                +
                +

                Group-by notes +

                The disk.frame implements the chunk_group_by operation with a significant caveat. In the disk.frame framework, group-by happens WITHIN each chunk and not ACROSS chunks. To achieve group by across chunk we need to put all rows with the same group keys into the same file chunk; this can be achieved with hard_group_by. However, the hard_group_by operation can be VERY TIME CONSUMING computationally and should be avoided if possible.

                The hard_group_by operation is best illustrated with an example, suppose a disk.frame has three chunks

                -
                # chunk1 = 1.fst
-#  id n
-#1  a 1
-#2  a 2
-#3  b 3
-#4  d 4
+
                # chunk1 = 1.fst
+#  id n
+#1  a 1
+#2  a 2
+#3  b 3
+#4  d 4
 
-# chunk2 = 2.fst
-#  id n
-#1  a 4
-#2  a 5
-#3  b 6
-#4  d 7
+# chunk2 = 2.fst
+#  id n
+#1  a 4
+#2  a 5
+#3  b 6
+#4  d 7
 
-# chunk3 = 3.fst
-#  id n
-#1  a 4
-#2  b 5
-#3  c 6
                
-
                and notice that the id column contains 3 distinct values "a","b", and "c". To perform hard_group_by(df, by = id) MAY give you the following disk.frame where all the ids with the same values end up in the same chunks.
                
-
                # chunk1 = 1.fst
-#  id n
-#1  b 3
-#2  b 6
+# chunk3 = 3.fst
+#  id n
+#1  a 4
+#2  b 5
+#3  c 6
                
+
                and notice that the id column contains 3 distinct values "a","b", and "c". To perform hard_group_by(df, by = id) MAY give you the following disk.frame where all the ids with the same values end up in the same chunks.
                
+
                # chunk1 = 1.fst
+#  id n
+#1  b 3
+#2  b 6
 
-# chunk2 = 2.fst
-#  id n
-#1  c 6
-#2  d 4
-#3  d 7
+# chunk2 = 2.fst
+#  id n
+#1  c 6
+#2  d 4
+#3  d 7
 
-# chunk3 = 3.fst
-#  id n
-#1  a 1
-#2  a 2
-#3  a 4
-#4  a 5
-#5  a 4
                
+# chunk3 = 3.fst
+#  id n
+#1  a 1
+#2  a 2
+#3  a 4
+#4  a 5
+#5  a 4

                Also, notice that there is no guaranteed order for the distribution of the ids to the chunks. The order is random, but each chunk is likely to have a similar number of rows, provided that id does not follow a skewed distribution i.e. where a few distinct values make up the majority of the rows.

                Typically, chunk_group_by is performed WITHIN each chunk. This is not an issue if the chunks have already been sharded on the by variables beforehand; however, if this is not the case then one may need a second stage aggregation to obtain the correct result, see Two-stage group by.

                By forcing the user to choose chunk_group_by (within each chunk) and hard_group_by (across all chunks), this ensures that the user is conscious of the choice they are making. In sparklyr the equivalent of a hard_group_by is performed, which we should avoid, where possible, as it is time-consuming and expensive. Hence, disk.frame has chosen to explain the theory and allow the user to make a conscious choice when performing group_by.

                -suppressMessages(library(disk.frame))
-flights.df %>%
-  hard_group_by(carrier) %>% # notice that hard_group_by needs to be set
-  chunk_summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>%  # mean follows normal R rules
-  collect %>% 
-  arrange(carrier)
                +suppressMessages(library(disk.frame)) +flights.df %>% + hard_group_by(carrier) %>% # notice that hard_group_by needs to be set + chunk_summarize(count = n(), mean_dep_delay = mean(dep_delay, na.rm=T)) %>% # mean follows normal R rules + collect %>% + arrange(carrier)
                -
                -

                -Two-Stage Group-by

                -

                Prior to disk.frame v0.3.0, there is no general support for one-stage group-by. Hence a two-stage style group-by is needed. The key is understand is the chunk_group_by which performs group-by within each chunk.

                +
                +

                Two-Stage Group-by +

                +

                Prior to disk.frame v0.3.0, there is no general support for one-stage group-by. Hence a two-stage style group-by is needed. The key is understand is the chunk_group_by which performs group-by within each chunk.

                For most group-by tasks, the user can achieve the desired result WITHOUT using hard = TRUE by performing the group by in two stages. For example, suppose you aim to count the number of rows group by carrier, you can set hard = F to find the count within each chunk and then use a second group-by to summaries each chunk’s results into the desired result. For example,

                -flights.df %>%
-  chunk_group_by(carrier) %>% # `chunk_group_by` aggregates within each chunk
-  chunk_summarize(count = n()) %>%  # mean follows normal R rules
-  collect %>%  # collect each individul chunks results and row-bind into a data.table
-  group_by(carrier) %>% 
-  summarize(count = sum(count)) %>% 
-  arrange(carrier)
                +flights.df %>% + chunk_group_by(carrier) %>% # `chunk_group_by` aggregates within each chunk + chunk_summarize(count = n()) %>% # mean follows normal R rules + collect %>% # collect each individul chunks results and row-bind into a data.table + group_by(carrier) %>% + summarize(count = sum(count)) %>% + arrange(carrier)

                Because this two-stage approach avoids the expensive hard group_by operation, it is often significantly faster. However, it can be tedious to write; and this is a con of the disk.frame chunking mechanism.

                Note: this two-stage approach is similar to a map-reduce operation.

                -flights.df = as.disk.frame(nycflights13::flights)
+flights.df = as.disk.frame(nycflights13::flights)
 
-flights.df %>%
-  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
-  chunk_group_by(year) %>% 
-  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
+flights.df %>%
+  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
+  chunk_group_by(year) %>% 
+  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
   collect

                This is two-stage group-by in action

                 # need a 2nd stage to finalise summing
-flights.df %>%
-  srckeep(c("year","distance")) %>%  # keep only carrier and distance columns
-  chunk_group_by(year) %>% 
-  chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk
-  collect %>% 
-  group_by(year) %>% 
-  summarise(sum_dist = sum(sum_dist))
                +flights.df %>% + srckeep(c("year","distance")) %>% # keep only carrier and distance columns + chunk_group_by(year) %>% + chunk_summarise(sum_dist = sum(distance)) %>% # this does a count per chunk + collect %>% + group_by(year) %>% + summarise(sum_dist = sum(sum_dist))

                You can mix group-by with other dplyr verbs as below, here is an example of using filter.

                 # filter
-pt = proc.time()
+pt = proc.time()
 df_filtered <-
-  flights.df %>% 
-  filter(month == 1)
-cat("filtering a < 0.1 took: ", data.table::timetaken(pt), "\n")
-nrow(df_filtered)
                + flights.df %>% + filter(month == 1) +cat("filtering a < 0.1 took: ", data.table::timetaken(pt), "\n") +nrow(df_filtered)
              • -
              -

              -Hard group-by

              +
              +

              Hard group-by +

              Another way to perform a one-stage group_by is to perform a hard_group_by on a disk.frame. This will rechunk the disk.frame by the by-columns. This is not recommended for performance reasons, as it can be quite slow to rechunk the file chunks on disk.

              -pt = proc.time()
-res1 <- flights.df %>% 
-  srckeep(c("month", "dep_delay")) %>% 
-  filter(month <= 6) %>% 
-  mutate(qtr = ifelse(month <= 3, "Q1", "Q2")) %>% 
-  hard_group_by(qtr) %>% # hard group_by is MUCH SLOWER but avoid a 2nd stage aggregation
-  chunk_summarise(avg_delay = mean(dep_delay, na.rm = TRUE)) %>% 
+pt = proc.time()
+res1 <- flights.df %>% 
+  srckeep(c("month", "dep_delay")) %>% 
+  filter(month <= 6) %>% 
+  mutate(qtr = ifelse(month <= 3, "Q1", "Q2")) %>% 
+  hard_group_by(qtr) %>% # hard group_by is MUCH SLOWER but avoid a 2nd stage aggregation
+  chunk_summarise(avg_delay = mean(dep_delay, na.rm = TRUE)) %>% 
   collect
-cat("group-by took: ", data.table::timetaken(pt), "\n")
+cat("group-by took: ", data.table::timetaken(pt), "\n")
 
-collect(res1)
              +collect(res1)
              @@ -385,11 +380,13 @@

              -

              Developed by Dai ZJ.

              +

              +

              Developed by Dai ZJ.

              -

              Site built with pkgdown 1.6.1.

              +

              +

              Site built with pkgdown 2.0.2.

              @@ -398,5 +395,7 @@

              + + diff --git a/docs/articles/11-custom-group-by.html b/docs/articles/11-custom-group-by.html index 53eb0bf7..356dfa28 100644 --- a/docs/articles/11-custom-group-by.html +++ b/docs/articles/11-custom-group-by.html @@ -19,6 +19,8 @@ + +
              +

              Custom One-Stage Group-by functions

              -

              ZJ

              +

              ZJ

              - Source: vignettes/11-custom-group-by.Rmd + Source: vignettes/11-custom-group-by.Rmd
              -
              -

              -Custom one-stage group-by

              +
              +

              Custom one-stage group-by +

              Group-by is one the most useful and commonly used operations in data manipulation. The most common to perform group with dplyr syntax would be something similar to the below

              -dataf %>% 
-  group_by(grpA) %>% 
-  summarize(mean(X))
              -

              However, the above nice and concise group-by syntax creates a challenge in disk.frame, because disk.frame organises the data in chunks and these chunks are often manipulated independently to each other which makes it difficult to compute the mean of any column if the group-by column is stored in different chunks. In this chapter, I will explain how disk.frame achieve the one-stage group-by syntax that you see, and how you can define custom one-stage group-by functions for use with disk.frame.

              -
              -

              -At a glance

              -

              disk.frame allows the user to create custom one-stage group-by functions. To make a function fn one stage, one needs to define two functions:

              +dataf %>% + group_by(grpA) %>% + summarize(mean(X))
              +

              However, the above nice and concise group-by syntax creates a challenge in disk.frame, because disk.frame organises the data in chunks and these chunks are often manipulated independently to each other which makes it difficult to compute the mean of any column if the group-by column is stored in different chunks. In this chapter, I will explain how disk.frame achieve the one-stage group-by syntax that you see, and how you can define custom one-stage group-by functions for use with disk.frame.

              +
              +

              At a glance +

              +

              disk.frame allows the user to create custom one-stage group-by functions. To make a function fn one stage, one needs to define two functions:

              1. fn_df.chunk_agg.disk.frame which applies itself to each chunk
              2. fn_df.collected_agg.disk.frame which accepts a list of returns from fn_df.chunk_agg.disk.frame and finalize the computation
              -

              For example, to make mean a one-stage group-by function, disk.frame has defined mean_df.chunk_agg.disk.frame and mean_df.collected_agg.disk.frame, which we will illustrate with examples below.

              -

              But first, we shall explain some theory behind disk.frame to help you better understand “why does disk.frame do it like that?”.

              +

              For example, to make mean a one-stage group-by function, disk.frame has defined mean_df.chunk_agg.disk.frame and mean_df.collected_agg.disk.frame, which we will illustrate with examples below.

              +

              But first, we shall explain some theory behind disk.frame to help you better understand “why does disk.frame do it like that?”.

              -
              -

              -How does {disk.frame} work

              -

              One may ask, how come only a few functions are supported for one-stage group-by? And why are some functions like median only produce estimates instead of producing the exact figure? To answer these questions, we need to have an understanding of how disk.frame works.

              -

              A disk.frame is organized as chunks stored on disk. Each chunk is a file stored in fst format. The {future} package is used to apply the same function to each chunk, each of these operations are carried out in a separate R session. These R sessions cannot communicate with each other during the execution of the operations.

              +
              +

              How does {disk.frame} work +

              +

              One may ask, how come only a few functions are supported for one-stage group-by? And why are some functions like median only produce estimates instead of producing the exact figure? To answer these questions, we need to have an understanding of how disk.frame works.

              +

              A disk.frame is organized as chunks stored on disk. Each chunk is a file stored in fst format. The {future} package is used to apply the same function to each chunk, each of these operations are carried out in a separate R session. These R sessions cannot communicate with each other during the execution of the operations.

              Once the operations have been performed on all chunks, the results will be bought back to the session from which the operations were called. This is the only point of inter-process communication.

              To summarize, the two phases of a df %>% some_fn %>% collect operation are

                @@ -161,14 +157,14 @@

                collect then row-binds (rbind/bind_rows/rbindlist) the results together to form a data.frame in the main session

              -
              -

              -How group-by works

              -

              Except for passing the result back to the main session, communication between worker sessions are not allowed. This limits how group-by operations can be performed, hence why group-by was done in two stages prior to disk.frame v0.3.0. However, R’s meta-programming abilities allow us to rewrite code to perform the two-stage group-bys using one-stage group-by code. For example, consider:

              +
              +

              How group-by works +

              +

              Except for passing the result back to the main session, communication between worker sessions are not allowed. This limits how group-by operations can be performed, hence why group-by was done in two stages prior to disk.frame v0.3.0. However, R’s meta-programming abilities allow us to rewrite code to perform the two-stage group-bys using one-stage group-by code. For example, consider:

              -df %>% 
-  group_by(grp1) %>% 
-  summarize(sum(x)) %>% 
+df %>% 
+  group_by(grp1) %>% 
+  summarize(sum(x)) %>% 
   collect

              we can use meta-programming to transform that to

              df %>% 
@@ -179,11 +175,11 @@ 
              
   summarize(x = sum(__tmp1__))

              Basically, we are “compiling” one-stage group-by code to two-stage group-by code, and then executing it.

              For mean, it’s trickier, as one needs to keep track on the numerator and the denominator separately in computing mean(x) = sum(x)/length(x).

              -

              Therefore, disk.frame compiles

              +

              Therefore, disk.frame compiles

              -df %>% 
-  group_by(grp1) %>% 
-  summarize(meanx = mean(x)) %>% 
+df %>% 
+  group_by(grp1) %>% 
+  summarize(meanx = mean(x)) %>% 
   collect

              to

              df %>% 
@@ -196,27 +192,27 @@ 
              
 
              Here is an example implementation of mean.chunk_agg.disk.frame
              
 
               mean_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE, ...) {
-  sumx = sum(x, na.rm = na.rm)
-  lengthx = length(x) - ifelse(na.rm, sum(is.na(x)), 0)
-  c(sumx = sumx, lengthx = lengthx)
+  sumx = sum(x, na.rm = na.rm)
+  lengthx = length(x) - ifelse(na.rm, sum(is.na(x)), 0)
+  c(sumx = sumx, lengthx = lengthx)
 }
              
 
              The mean_df.collected_agg.disk.frame receives a list of outputs from mean_df.chunk_agg.disk.frame. Recall that mean.chunk_agg.disk.frame returns a vector for each chunk, so the input to mean_df.collected_agg.disk.frame is a list of vectors
              
 
               mean_df.collected_agg.disk.frame <- function(listx) {
-  sum(sapply(listx, function(x) x["sumx"]))/sum(sapply(listx, function(x) x["lengthx"]))
+  sum(sapply(listx, function(x) x["sumx"]))/sum(sapply(listx, function(x) x["lengthx"]))
 }
              -
              -

              -How to define custom one-stage group-by functions

              +
              +

              How to define custom one-stage group-by functions +

              Now that we have seen two examples, namely sum and mean, we are ready to summarize how group-by functions are implemented.

              Given the below

              -df %>% 
-  group_by(grp1) %>% 
-  summarize(namex = fn(x)) %>% 
+df %>% 
+  group_by(grp1) %>% 
+  summarize(namex = fn(x)) %>% 
   collect
              -

              disk.frame compiles it to

              +

              disk.frame compiles it to

              df %>% 
   chunk_group_by(grp1) %>% 
   chunk_summarize(__tmp1__ = list(fn_df.chunk_agg.disk.frame(x))) %>% 
@@ -236,26 +232,26 @@ 
              
 
 
               sum_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE) {
-  sum(x, na.rm=na.rm)
+  sum(x, na.rm=na.rm)
 }
              
 
                
-
              1. Define sum_df.collected_agg.disk.frame, which needs to accept a list of sum(x, na.rm), but sum(x, na.rm) is just a numeric, so
                2. 
+
              2. Define sum_df.collected_agg.disk.frame, which needs to accept a list of sum(x, na.rm), but sum(x, na.rm) is just a numeric, so
                3. 
 
              
 
               sum_df.collected_agg.disk.frame <- function(list_sum) {
-  sum(unlist(list_sum))
+  sum(unlist(list_sum))
 }
              
 
              Example of implementing n_distinct:
              
-
              The dplyr::n_distinct function counts the number of distint values from a vector x
              
+
              The dplyr::n_distinct function counts the number of distint values from a vector x
              
 
                
 
              1. Define n_distinct_df.chunk_agg.disk.frame, to return a list of unique values. Because the same value can appear in multiple chunks, to ensure that we don’t double count, we simply return all the unique values from each chunk which is then de-duplicated in the next phase
                2. 
 
              
 
               n_distinct_df.chunk_agg.disk.frame <- function(x, na.rm = FALSE) {
   if(na.rm) {
-    setdiff(unique(x), NA)
+    setdiff(unique(x), NA)
   } else {
-    unique(x)
+    unique(x)
   }
 }
              
 
                
@@ -263,20 +259,20 @@ 
                
 
              
 
               n_distinct_df.collected_agg.disk.frame <- function(list_of_chunkwise_uniques) {
-  dplyr::n_distinct(unlist(list_of_chunkwise_uniques))
+  dplyr::n_distinct(unlist(list_of_chunkwise_uniques))
 }
              -
              -

              -Limitations

              -

              We have seen that disk.frame performs operations in two phases

              +
              +

              Limitations +

              +

              We have seen that disk.frame performs operations in two phases

              1. apply the same function to each chunk
              2. row-bind the results

              and there are no communication between the sessions that applies the functions at chunk level.

              Hence, it is generally difficult to compute rank based summarizations like median exactly. Hence most rank based calculations are estimates only. This is also true of distributed data system like Spark whose median function is also estimates only.

              -

              Another limitation for now is that summarization that is more complex then f(x) is not supported. E.g. sum(x) + 1, sum(x + mean(x)), sum(x) + mean(x), and fn(sum(x)) are not yet supported as arguments in the summarize function.

              +

              Another limitation for now is that summarization that is more complex then f(x) is not supported. E.g. sum(x) + 1, sum(x + mean(x)), sum(x) + mean(x), and fn(sum(x)) are not yet supported as arguments in the summarize function.

              @@ -292,11 +288,13 @@

              -

              Developed by Dai ZJ.

              +

              +

              Developed by Dai ZJ.

              -

              Site built with pkgdown 1.6.1.

              +

              +

              Site built with pkgdown 2.0.2.

              @@ -305,5 +303,7 @@

              + + diff --git a/docs/articles/88-trouble-shooting.html b/docs/articles/88-trouble-shooting.html index 336e9c5b..ab2a72a9 100644 --- a/docs/articles/88-trouble-shooting.html +++ b/docs/articles/88-trouble-shooting.html @@ -19,6 +19,8 @@ + +
              +

              Trouble shooting

              -

              ZJ

              +

              ZJ

              - Source: vignettes/88-trouble-shooting.Rmd + Source: vignettes/88-trouble-shooting.Rmd
              -
              -

              -Steps to trouble shoot

              +
              +

              Steps to trouble shoot +

              1. I suggest updating {future} and your R version if you have not already done so.

              2. Are you able to share the data?

              3. Do a good MWE

              -
              library(disk.frame)
-setup_disk.frame()
+
              library(disk.frame)
+setup_disk.frame()
 
-df<-as.disk.frame(a)
+df<-as.disk.frame(a)
 
 
-df1 = mutate(df, date = as.Date(as.character(datadate), format="%Y%m%d"))
+df1 = mutate(df, date = as.Date(as.character(datadate), format="%Y%m%d"))
 
-head(df1)
              
+head(df1)

              1. Check if your virus scanner is blocking interprocess communication

              2. Try to apply the function to just one chunk, perhaps there is a syntax error or column error? If one chunk works then you can rule out coding error

              -
              get_chunk(df, 1) %>%
-  mutate(date = as.Date(as.character(datadate), format="%Y%m%d"))
              +
              get_chunk(df, 1) %>%
+  mutate(date = as.Date(as.character(datadate), format="%Y%m%d"))
              1. Set the number of workers to 1, so there is no more inter-process communication. Does it work now? If it does, then it’s the inter process communication. You might need to contact your admin for help
              @@ -169,11 +165,13 @@

              -

              Developed by Dai ZJ.

              +

              +

              Developed by Dai ZJ.

              -

              Site built with pkgdown 1.6.1.

              +

              +

              Site built with pkgdown 2.0.2.

              @@ -182,5 +180,7 @@

              + + diff --git a/docs/articles/index.html b/docs/articles/index.html index 3569fe74..f767f6f8 100644 --- a/docs/articles/index.html +++ b/docs/articles/index.html @@ -1,66 +1,12 @@ - - - - - - - -Articles • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Articles • disk.frame - + + - - - -
              -
              -

            • - - -
            +

            Articles

            @@ -162,53 +93,49 @@

            Articles

            All vignettes

            -
            -
            Preface - The birth of `disk.frame`
            -
            -
            Quick Start: Basic Operations with nycflights13
            -
            -
            Key `{disk.frame}` concepts
            -
            -
            Ingesting Data
            -
            -
            Using data.table syntax with disk.frame
            -
            -
            Benchmarks 1: disk.frame beats Dask! disk.frame beats JuliaDB! Anyone else wanna challenge?
            -
            -
            Generalized Linear Models (GLM) including logistic regression with disk.frame
            -
            -
            {disk.frame} can be more 'epic'
            -
            -
            Convenience features
            -
            -
            Group-by
            -
            -
            Custom One-Stage Group-by functions
            -
            -
            Trouble shooting
            -
            -
            -
            +
            Preface - The birth of `disk.frame`
            +
            +
            Quick Start: Basic Operations with nycflights13
            +
            +
            Key `{disk.frame}` concepts
            +
            +
            Ingesting Data
            +
            +
            Using data.table syntax with disk.frame
            +
            +
            Benchmarks 1: disk.frame beats Dask! disk.frame beats JuliaDB! Anyone else wanna challenge?
            +
            +
            Generalized Linear Models (GLM) including logistic regression with disk.frame
            +
            +
            {disk.frame} can be more 'epic'
            +
            +
            Convenience features
            +
            +
            Group-by
            +
            +
            Custom One-Stage Group-by functions
            +
            +
            Trouble shooting
            +
            +
            -
            -
            -

            Developed by Dai ZJ.

            +
            +

            Developed by Dai ZJ.

            -

            Site built with pkgdown 1.6.1.

            +

            Site built with pkgdown 2.0.2.

            -
            -
            +
          • - - + + diff --git a/docs/authors.html b/docs/authors.html index a2bdd37f..342af2fa 100644 --- a/docs/authors.html +++ b/docs/authors.html @@ -1,66 +1,12 @@ - - - - - - - -Authors • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Authors and Citation • disk.frame + + - - - - - -
          -
          -
        - - -
        +
        -
        -

        Authors

        +
        +
        +

        Authors

        +
        + + +
        • +

          Dai ZJ. Author, maintainer. +

          +
          • +
        • +

          Jacky Poon. Contributor. +

          +
          • +
        +
        +
        +

        Citation

        + Source: DESCRIPTION +
        -
          -
        • -

          Dai ZJ. Author, maintainer. -

          -
          • -
        • -

          Jacky Poon. Contributor. -

          -
          • -
        + +

        ZJ D (2022). +disk.frame: Larger-than-RAM Disk-Based Data Manipulation Framework. +R package version 0.5.0, https://diskframe.com. +

        + 
        @Manual{,
+  title = {disk.frame: Larger-than-RAM Disk-Based Data Manipulation Framework},
+  author = {Dai ZJ},
+  year = {2022},
+  note = {R package version 0.5.0},
+  url = {https://diskframe.com},
+}
        @@ -175,22 +126,20 @@

        Authors

        -
        -
        -

        Developed by Dai ZJ.

        +
        +

        Developed by Dai ZJ.

        -

        Site built with pkgdown 1.6.1.

        +

        Site built with pkgdown 2.0.2.

        -
        -
        +
        - - + + diff --git a/docs/index.html b/docs/index.html index d63f78e2..89c4a22e 100644 --- a/docs/index.html +++ b/docs/index.html @@ -23,6 +23,8 @@ + +
        -
        -

        -disk.frame +
        +

        disk.frame

        -
        -

        -Introduction

        +
        +

        Introduction +

        How do I manipulate tabular data that doesn’t fit into Random Access Memory (RAM)?

        -

        Use disk.frame!

        -

        In a nutshell, disk.frame makes use of two simple ideas

        -
          +

          Use disk.frame!

          +

          In a nutshell, disk.frame makes use of two simple ideas

          +
          1. split up a larger-than-RAM dataset into chunks and store each chunk in a separate file inside a folder and
          2. provide a convenient API to manipulate these chunks
          -

          disk.frame performs a similar role to distributed systems such as Apache Spark, Python’s Dask, and Julia’s JuliaDB.jl for medium data which are datasets that are too large for RAM but not quite large enough to qualify as big data.

          -
          -

          -Installation

          -

          You can install the released version of disk.frame from CRAN with:

          +

          disk.frame performs a similar role to distributed systems such as Apache Spark, Python’s Dask, and Julia’s JuliaDB.jl for medium data which are datasets that are too large for RAM but not quite large enough to qualify as big data.

          +
          +

          Installation +

          +

          You can install the released version of disk.frame from CRAN with:

          -install.packages("disk.frame")
          -

          And the development version from GitHub with:

          +install.packages("disk.frame")
          +

          And the development version from GitHub with:

           # install.packages("devtools")
-devtools::install_github("xiaodaigh/disk.frame")
          +devtools::install_github("xiaodaigh/disk.frame")

          On some platforms, such as SageMaker, you may need to explicitly specify a repo like this

          -install.packages("disk.frame", repo="https://cran.rstudio.com")
          +install.packages("disk.frame", repo="https://cran.rstudio.com")
        -
        -

        -Vignettes and articles

        -

        Please see these vignettes and articles about disk.frame

        +
        +

        Vignettes and articles +

        +

        Please see these vignettes and articles about disk.frame

        -
        -

        -Common questions

        -
        -

        -a) What is {disk.frame} and why create it?

        -

        disk.frame is an R package that provides a framework for manipulating larger-than-RAM structured tabular data on disk efficiently. The reason one would want to manipulate data on disk is that it allows arbitrarily large datasets to be processed by R. In other words, we go from “R can only deal with data that fits in RAM” to “R can deal with any data that fits on disk”. See the next section.

        -
        -
        -

        -b) How is it different to data.frame and data.table?

        +
        +

        Common questions +

        +
        +

        a) What is {disk.frame} and why create it? +

        +

        disk.frame is an R package that provides a framework for manipulating larger-than-RAM structured tabular data on disk efficiently. The reason one would want to manipulate data on disk is that it allows arbitrarily large datasets to be processed by R. In other words, we go from “R can only deal with data that fits in RAM” to “R can deal with any data that fits on disk”. See the next section.

        +
        +
        +

        b) How is it different to data.frame and data.table? +

        A data.frame in R is an in-memory data structure, which means that R must load the data in its entirety into RAM. A corollary of this is that only data that can fit into RAM can be processed using data.frames. This places significant restrictions on what R can process with minimal hassle.

        -

        In contrast, disk.frame provides a framework to store and manipulate data on the hard drive. It does this by loading only a small part of the data, called a chunk, into RAM; process the chunk, write out the results and repeat with the next chunk. This chunking strategy is widely applied in other packages to enable processing large amounts of data in R, for example, see chunkded arkdb, and iotools.

        -

        Furthermore, there is a row-limit of 2^31 for data.frames in R; hence an alternate approach is needed to apply R to these large datasets. The chunking mechanism in disk.frame provides such an avenue to enable data manipulation beyond the 2^31 row limit.

        -
        -
        -

        -c) How is {disk.frame} different to previous “big” data solutions for R?

        -

        R has many packages that can deal with larger-than-RAM datasets, including ff and bigmemory. However, ff and bigmemory restrict the user to primitive data types such as double, which means they do not support character (string) and factor types. In contrast, disk.frame makes use of data.table::data.table and data.frame directly, so all data types are supported. Also, disk.frame strives to provide an API that is as similar to data.frame’s where possible. disk.frame supports many dplyr verbs for manipulating disk.frames.

        -

        Additionally, disk.frame supports parallel data operations using infrastructures provided by the excellent future package to take advantage of multi-core CPUs. Further, disk.frame uses state-of-the-art data storage techniques such as fast data compression, and random access to rows and columns provided by the fst package to provide superior data manipulation speeds.

        -
        -
        -

        -d) How does {disk.frame} work?

        -

        disk.frame works by breaking large datasets into smaller individual chunks and storing the chunks in fst files inside a folder. Each chunk is a fst file containing a data.frame/data.table. One can construct the original large dataset by loading all the chunks into RAM and row-bind all the chunks into one large data.frame. Of course, in practice this isn’t always possible; hence why we store them as smaller individual chunks.

        -

        disk.frame makes it easy to manipulate the underlying chunks by implementing dplyr functions/verbs and other convenient functions (e.g. the cmap(a.disk.frame, fn, lazy = F) function which applies the function fn to each chunk of a.disk.frame in parallel). So that disk.frame can be manipulated in a similar fashion to in-memory data.frames.

        -
        -
        -

        -e) How is {disk.frame} different from Spark, Dask, and JuliaDB.jl?

        -

        Spark is primarily a distributed system that also works on a single machine. Dask is a Python package that is most similar to disk.frame, and JuliaDB.jl is a Julia package. All three can distribute work over a cluster of computers. However, disk.frame currently cannot distribute data processes over many computers, and is, therefore, single machine focused.

        -

        In R, one can access Spark via sparklyr, but that requires a Spark cluster to be set up. On the other hand disk.frame requires zero-setup apart from running install.packages("disk.frame") or devtools::install_github("xiaodaigh/disk.frame").

        -

        Finally, Spark can only apply functions that are implemented for Spark, whereas disk.frame can use any function in R including user-defined functions.

        -
        -
        -
        -
        -

        -Example usage

        -
        -

        -Set-up {disk.frame} +

        In contrast, disk.frame provides a framework to store and manipulate data on the hard drive. It does this by loading only a small part of the data, called a chunk, into RAM; process the chunk, write out the results and repeat with the next chunk. This chunking strategy is widely applied in other packages to enable processing large amounts of data in R, for example, see chunkded arkdb, and iotools.

        +

        Furthermore, there is a row-limit of 2^31 for data.frames in R; hence an alternate approach is needed to apply R to these large datasets. The chunking mechanism in disk.frame provides such an avenue to enable data manipulation beyond the 2^31 row limit.

        +

        +
        +

        c) How is {disk.frame} different to previous “big” data solutions for R? +

        +

        R has many packages that can deal with larger-than-RAM datasets, including ff and bigmemory. However, ff and bigmemory restrict the user to primitive data types such as double, which means they do not support character (string) and factor types. In contrast, disk.frame makes use of data.table::data.table and data.frame directly, so all data types are supported. Also, disk.frame strives to provide an API that is as similar to data.frame’s where possible. disk.frame supports many dplyr verbs for manipulating disk.frames.

        +

        Additionally, disk.frame supports parallel data operations using infrastructures provided by the excellent future package to take advantage of multi-core CPUs. Further, disk.frame uses state-of-the-art data storage techniques such as fast data compression, and random access to rows and columns provided by the fst package to provide superior data manipulation speeds.

        +
        +
        +

        d) How does {disk.frame} work? +

        +

        disk.frame works by breaking large datasets into smaller individual chunks and storing the chunks in fst files inside a folder. Each chunk is a fst file containing a data.frame/data.table. One can construct the original large dataset by loading all the chunks into RAM and row-bind all the chunks into one large data.frame. Of course, in practice this isn’t always possible; hence why we store them as smaller individual chunks.

        +

        disk.frame makes it easy to manipulate the underlying chunks by implementing dplyr functions/verbs and other convenient functions (e.g. the cmap(a.disk.frame, fn, lazy = F) function which applies the function fn to each chunk of a.disk.frame in parallel). So that disk.frame can be manipulated in a similar fashion to in-memory data.frames.

        +
        +
        +

        e) How is {disk.frame} different from Spark, Dask, and JuliaDB.jl? +

        +

        Spark is primarily a distributed system that also works on a single machine. Dask is a Python package that is most similar to disk.frame, and JuliaDB.jl is a Julia package. All three can distribute work over a cluster of computers. However, disk.frame currently cannot distribute data processes over many computers, and is, therefore, single machine focused.

        +

        In R, one can access Spark via sparklyr, but that requires a Spark cluster to be set up. On the other hand disk.frame requires zero-setup apart from running install.packages("disk.frame") or devtools::install_github("xiaodaigh/disk.frame").

        +

        Finally, Spark can only apply functions that are implemented for Spark, whereas disk.frame can use any function in R including user-defined functions.

        +
        +
        +
        +
        +

        Example usage +

        +
        +

        Set-up {disk.frame} +

        -

        disk.frame works best if it can process multiple data chunks in parallel. The best way to set-up disk.frame so that each CPU core runs a background worker is by using

        +

        disk.frame works best if it can process multiple data chunks in parallel. The best way to set-up disk.frame so that each CPU core runs a background worker is by using

         setup_disk.frame()
 
 # this allows large datasets to be transferred between sessions
-options(future.globals.maxSize = Inf)
        +options(future.globals.maxSize = Inf)

        The setup_disk.frame() sets up background workers equal to the number of CPU cores; please note that, by default, hyper-threaded cores are counted as one not two.

        -

        Alternatively, one may specify the number of workers using setup_disk.frame(workers = n).

        +

        Alternatively, one may specify the number of workers using setup_disk.frame(workers = n).

        -
        -

        -Quick-start

        +
        +

        Quick-start +

        -suppressPackageStartupMessages(library(disk.frame))
-library(nycflights13)
+suppressPackageStartupMessages(library(disk.frame))
+library(nycflights13)
 
 # this will setup disk.frame's parallel backend with number of workers equal to the number of CPU cores (hyper-threaded cores are counted as one not two)
 setup_disk.frame()
 #> The number of workers available for disk.frame is 6
 # this allows large datasets to be transferred between sessions
-options(future.globals.maxSize = Inf)
+options(future.globals.maxSize = Inf)
 
 # convert the flights data.frame to a disk.frame
 # optionally, you may specify an outdir, otherwise, the 
-flights.df <- as.disk.frame(nycflights13::flights)
        -
        -
        -

        -Example: dplyr verbs

        -
        -

        -dplyr verbs

        +flights.df <- as.disk.frame(nycflights13::flights)
        +
        +
        +

        Example: dplyr verbs +

        +
        +

        dplyr verbs +

        {disk.frame} aims to support as many dplyr verbs as possible. For example

        -flights.df %>% 
-  filter(year == 2013) %>% 
-  mutate(origin_dest = paste0(origin, dest)) %>% 
-  head(2)
-#>   year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier flight
-#> 1 2013     1   1      517            515         2      830            819        11      UA   1545
-#> 2 2013     1   1      533            529         4      850            830        20      UA   1714
-#>   tailnum origin dest air_time distance hour minute           time_hour origin_dest
-#> 1  N14228    EWR  IAH      227     1400    5     15 2013-01-01 05:00:00      EWRIAH
-#> 2  N24211    LGA  IAH      227     1416    5     29 2013-01-01 05:00:00      LGAIAH
        -
        -
        -

        -Group-by

        -

        Starting from disk.frame v0.3.0, there is group_by support for a limited set of functions. For example:

        +flights.df %>% + filter(year == 2013) %>% + mutate(origin_dest = paste0(origin, dest)) %>% + head(2) +#> year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier +#> 1 2013 1 1 517 515 2 830 819 11 UA +#> 2 2013 1 1 533 529 4 850 830 20 UA +#> flight tailnum origin dest air_time distance hour minute time_hour origin_dest +#> 1 1545 N14228 EWR IAH 227 1400 5 15 2013-01-01 05:00:00 EWRIAH +#> 2 1714 N24211 LGA IAH 227 1416 5 29 2013-01-01 05:00:00 LGAIAH
        +
        +
        +

        Group-by +

        +

        Starting from disk.frame v0.3.0, there is group_by support for a limited set of functions. For example:

        -result_from_disk.frame = iris %>% 
-  as.disk.frame %>% 
-  group_by(Species) %>% 
-  summarize(
-    mean(Petal.Length), 
-    sumx = sum(Petal.Length/Sepal.Width), 
-    sd(Sepal.Width/ Petal.Length), 
-    var(Sepal.Width/ Sepal.Width), 
-    l = length(Sepal.Width/ Sepal.Width + 2),
-    max(Sepal.Width), 
-    min(Sepal.Width), 
-    median(Sepal.Width)
-    ) %>% 
+result_from_disk.frame = iris %>% 
+  as.disk.frame %>% 
+  group_by(Species) %>% 
+  summarize(
+    mean(Petal.Length), 
+    sumx = sum(Petal.Length/Sepal.Width), 
+    sd(Sepal.Width/ Petal.Length), 
+    var(Sepal.Width/ Sepal.Width), 
+    l = length(Sepal.Width/ Sepal.Width + 2),
+    max(Sepal.Width), 
+    min(Sepal.Width), 
+    median(Sepal.Width)
+    ) %>% 
   collect
        -

        The results should be exactly the same as if applying the same group-by operations on a data.frame. If not, please report a bug.

        -
        -

        -List of supported group-by functions

        -

        If a function you like is missing, please make a feature request here. It is a limitation that function that depend on the order a column can only be obtained using estimated methods.

        +

        The results should be exactly the same as if applying the same group-by operations on a data.frame. If not, please report a bug.

        +
        +

        List of supported group-by functions +

        +

        If a function you like is missing, please make a feature request here. It is a limitation that function that depend on the order a column can only be obtained using estimated methods.

        @@ -335,7 +330,7 @@

        @@ -368,19 +363,19 @@

        -
        -

        -Example: data.table syntax

        +
        +

        Example: data.table syntax +

        -library(data.table)
+library(data.table)
 
-suppressWarnings(
+suppressWarnings(
   grp_by_stage1 <- 
     flights.df[
-      keep = c("month", "distance"), # this analysis only required "month" and "dist" so only load those
+      keep = c("month", "distance"), # this analysis only required "month" and "dist" so only load those
       month <= 6, 
-      .(sum_dist = sum(distance)), 
-      .(qtr = ifelse(month <= 3, "Q1", "Q2"))
+      .(sum_dist = sum(distance)), 
+      .(qtr = ifelse(month <= 3, "Q1", "Q2"))
       ]
 )
 
@@ -394,60 +389,62 @@ 
        
 #> 6:  Q2 27397926

        The result grp_by_stage1 is a data.table so we can finish off the two-stage aggregation using data.table syntax

        -grp_by_stage2 = grp_by_stage1[,.(sum_dist = sum(sum_dist)), qtr]
+grp_by_stage2 = grp_by_stage1[,.(sum_dist = sum(sum_dist)), qtr]
 
 grp_by_stage2
 #>    qtr sum_dist
 #> 1:  Q1 81343950
 #> 2:  Q2 89257810
        -
        -

        -Basic info

        +
        +

        Basic info +

        To find out where the disk.frame is stored on disk:

         # where is the disk.frame stored
-attr(flights.df, "path")
-#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpEdhlRv\\file33ec607f12cc.df"
        +attr(flights.df, "path") +#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\RtmpIlXNzn\\file568813b835a7.df"

        A number of data.frame functions are implemented for disk.frame

         # get first few rows
-head(flights.df, 1)
+head(flights.df, 1)
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier
 #> 1: 2013     1   1      517            515         2      830            819        11      UA
 #>    flight tailnum origin dest air_time distance hour minute           time_hour
 #> 1:   1545  N14228    EWR  IAH      227     1400    5     15 2013-01-01 05:00:00
         # get last few rows
-tail(flights.df, 1)
+tail(flights.df, 1)
 #>    year month day dep_time sched_dep_time dep_delay arr_time sched_arr_time arr_delay carrier
 #> 1: 2013     9  30       NA            840        NA       NA           1020        NA      MQ
 #>    flight tailnum origin dest air_time distance hour minute           time_hour
 #> 1:   3531  N839MQ    LGA  RDU       NA      431    8     40 2013-09-30 08:00:00
         # number of rows
-nrow(flights.df)
+nrow(flights.df)
 #> [1] 336776
         # number of columns
-ncol(flights.df)
+ncol(flights.df)
 #> [1] 19
        -

        Functionvar Exact -var(x) only cor, cov support planned +var(x) only cor, cov support planned
        @@ -476,67 +473,67 @@

        - + - - + - + - - + +disk.frame was number 3 - + - + - + +disk.frame used in helping to create a maximum likelihood estimation program for linear models - + - + - + @@ -544,67 +541,60 @@

        25 days of disk.frame25 days of disk.frame English ZJ 2019-12-0125 tweets about disk.frame +25 tweets about disk.frame
        https://www.researchgate.net/post/What-is-the-Maximum-size-of-data-that-is-supported-by-R-datamininghttps://www.researchgate.net/post/What-is-the-Maximum-size-of-data-that-is-supported-by-R-datamining English Knut Jägersberg 2019-11-11 Great answer on using disk.frame
        {disk.frame} is epic{disk.frame} is epic English Bruno Rodriguez 2019-09-03It’s about loading a 30G file into disk.frame +It’s about loading a 30G file into disk.frame
        My top 10 R packages for data analyticsMy top 10 R packages for data analytics English Jacky Poon 2019-09-03 -disk.frame was number 3
        useR! 2019 presentation videouseR! 2019 presentation video English Dai ZJ 2019-08-03
        useR! 2019 presentation slidesuseR! 2019 presentation slides English Dai ZJ 2019-08-03
        Split-apply-combine for Maximum Likelihood Estimation of a linear modelSplit-apply-combine for Maximum Likelihood Estimation of a linear model English Bruno Rodriguez 2019-10-06 -disk.frame used in helping to create a maximum likelihood estimation program for linear models
        Emma goes to useR! 2019Emma goes to useR! 2019 English Emma Vestesson 2019-07-16The first mention of disk.frame in a blog postThe first mention of disk.frame in a blog post
        深入对比数据科学工具箱：Python3 和 R 之争(2020版)深入对比数据科学工具箱：Python3 和 R 之争(2020版) Chinese Harry Zhu 2020-02-16
        -
        -

        -Interested in learning {disk.frame} in a structured course?

        +
        +

        Interested in learning {disk.frame} in a structured course? +

        Please register your interest at:

        -

        https://leanpub.com/c/taminglarger-than-ramwithdiskframe

        +

        https://leanpub.com/c/taminglarger-than-ramwithdiskframe

        -
        -

        -Open Collective

        -

        If you like disk.frame and want to speed up its development or perhaps you have a feature request? Please consider sponsoring disk.frame on Open Collective

        -
        -

        -Backers

        +
        +

        Open Collective +

        +

        If you like disk.frame and want to speed up its development or perhaps you have a feature request? Please consider sponsoring disk.frame on Open Collective

        +
        +

        Backers +

        Thank you to all our backers!

        -

        +

        - -
        -

        -Contact me for consulting

        +
        +

        Contact me for consulting +

        Do you need help with machine learning and data science in R, Python, or Julia? I am available for Machine Learning/Data Science/R/Python/Julia consulting! Email me

        -
        -

        -Non-financial ways to contribute

        +
        +

        Non-financial ways to contribute +

        Do you wish to give back the open-source community in non-financial ways? Here are some ways you can contribute

        - -
        -

        -Download Counts & Build Status

        -

        Travis build status

        +
        + +

        https://github.com/xiaodaigh/disk.frame-fannie-mae-example https://github.com/xiaodaigh/disk.frame-vs https://github.com/xiaodaigh/disk.frame.ml https://github.com/xiaodaigh/courses-larger-than-ram-data-manipulation-with-disk-frame

        -
        -

        -Live Stream of {disk.frame} development

        - +
        +

        Download Counts & Build Status +

        +

        @@ -612,42 +602,53 @@

        -

        Developed by Dai ZJ.

        +

        +

        Developed by Dai ZJ.

        -

        Site built with pkgdown 1.6.1.

        +

        +

        Site built with pkgdown 2.0.2.

        @@ -656,5 +657,7 @@

        Developers

        + + diff --git a/docs/news/index.html b/docs/news/index.html index f3b2ed7a..0fb5baf2 100644 --- a/docs/news/index.html +++ b/docs/news/index.html @@ -1,66 +1,12 @@ - - - - - - - -Changelog • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Changelog • disk.frame - - + + - - -
        -
        -

      - - -
      +

      Changelog

      - Source: NEWS.md + Source: NEWS.md
      -
      -

      -disk.frame 0.5 Unreleased -

      -
        -
      • removed add_count method
        • -
      -
      -
      -

      -disk.frame 0.4.1 2021-03-14 -

      -
        -
      • removed use of sysctl which was violating CRAN policy
        • -
      -
      -
      -

      -disk.frame 0.4.0 2021-02-14 -

      -
        -
      • Removed count and tally +
        +

        disk.frame 0.5

        +
        • removed add_count method
          • +
        +
        +

        disk.frame 0.4.12021-03-14

        +
        • removed use of sysctl which was violating CRAN policy
          • +
        +
        +

        disk.frame 0.4.02021-02-14

        +
        • Removed count and tally
        • Fixed package compatibility
          • -
        -
        -
        -

        -disk.frame 0.3.7 2020-07-07 -

        -
          -
        • Bugfix - add_chunk with date column not working
          • +
        +
        +

        disk.frame 0.3.72020-07-07

        +
        • Bugfix - add_chunk with date column not working
        • Minor - added compression and other aprameters to add_chunk
          • -
        -
        -
        -

        -disk.frame 0.3.6 2020-06-13 -

        -
          -
        • General update to dplyr v1
          • +
        +
        +

        disk.frame 0.3.62020-06-13

        +
        • General update to dplyr v1
        • Remove all _all/if/at functions for dplyr v1
        • Updated pull function to conform to dplyr
          • -
        -
        -
        -

        -disk.frame 0.3.5 2020-05-08 -

        - -
        -
        -

        -disk.frame 0.3.4 2020-02-26 -

        -
          -
        • fixed vignette bug and all vignette are now in static pdf
          • +
        +
        +

        disk.frame 0.3.52020-05-08

        +
        +
        +

        disk.frame 0.3.42020-02-26

        +
        • fixed vignette bug and all vignette are now in static pdf
        • fixed bugs with group-by when run with Rscript
        • also fixed bugs with multiple group-by columns
          • -
        -
        -
        -

        -disk.frame 0.3.3 2020-01-21 -

        -
          -
        • removed bloom filters as they were causing issues with CRAN solaris system for some reason
          • -
        -
        -
        -

        -disk.frame 0.3.2 2020-01-19 -

        -
          -
        • removed usage proc/meminfo inline with CRAN policy
          • +
        +
        +

        disk.frame 0.3.32020-01-21

        +
        • removed bloom filters as they were causing issues with CRAN solaris system for some reason
          • +
        +
        +

        disk.frame 0.3.22020-01-19

        +
        • removed usage proc/meminfo inline with CRAN policy
        • added (experimental) bloomfilter
          • -
        -
        -
        -

        -disk.frame 0.3.1 2019-12-20 -

        -
          -
        • urgent bug fix for group-by failing when the number chunks is 1; see Github #241
          • -
        -
        -
        -

        -disk.frame 0.3.0 2019-12-18 -

        -
          -
        • experimental one-stage group-by framework!
          • +
        +
        +

        disk.frame 0.3.12019-12-20

        +
        • urgent bug fix for group-by failing when the number chunks is 1; see Github #241
          • +
        +
        +

        disk.frame 0.3.02019-12-18

        +
        • experimental one-stage group-by framework!
        • bug fixes for data.table trigger by integration with tidyfast
        • removed assertthat from imports
        • add benchmarkme to Suggests
          • -
        -
        -
        -

        -disk.frame 0.2.1 2019-11-23 -

        -
          -
        • got rid of benchmarkme as a dependency
          • +
        +
        +

        disk.frame 0.2.12019-11-23

        +
        • got rid of benchmarkme as a dependency
        • added hard_arrange thanks to Jacky Poon
        • added more .progress options for joins
        • Using data.table::getDTthreads() as default number of workers
        • multisession instead of multiprocess as default backend for data.table
        • added support for R3.4
        • fixed df_get_ram for R < 3.6
          • -
        -
        -
        -

        -disk.frame 0.2.0 2019-10-05 -

        -
          -
        • deprecated group_by, arrange, summarise
          • +
        +
        +

        disk.frame 0.2.02019-10-05

        +
        • deprecated group_by, arrange, summarise
        • add chunk_group_by, chunk_arrange, chunk_summarise
        • fit GLMs with dfglm
        • fixed so that dplyr function also work in mutate even with ~ in the name
        • fixed disk.frame so that in works in functions too
          • -
        -
        -
        -

        -disk.frame 0.1.1 2019-09-15 -

        -
          -
        • Allowed map to accept multiple arguments. Thanks Knut Jägersberg for suggestion
          • +
        +
        +

        disk.frame 0.1.12019-09-15

        +
        • Allowed map to accept multiple arguments. Thanks Knut Jägersberg for suggestion
        • Fixed bug where if the CSV is larger than RAM then it fails by adding {LaF} backend
        • Added {bigreadr} backend for reading large files by first splitting the file. This is the default behaviour
        • Added {LaF} and {readr} chunk readers to csv_to_disk.frame
          • -
        • fixed write_disk.frame(...., shardby) and other shardby functions including rechunk and shard +
        • fixed write_disk.frame(...., shardby) and other shardby functions including rechunk and shard
        • added df_ram_size() to accurate determine RAM size for RStudio in R3.6
        • added show_ceremony and show_boiler_plate to show setup code
          • -
        -
        +
      +
      -
      -
      -

      Developed by Dai ZJ.

      +
      +

      Developed by Dai ZJ.

      -

      Site built with pkgdown 1.6.1.

      +

      Site built with pkgdown 2.0.2.

      -
      -
      +
      - - + + diff --git a/docs/pkgdown.css b/docs/pkgdown.css index 1273238d..80ea5b83 100644 --- a/docs/pkgdown.css +++ b/docs/pkgdown.css @@ -56,8 +56,10 @@ img.icon { float: right; } -img { +/* Ensure in-page images don't run outside their container */ +.contents img { max-width: 100%; + height: auto; } /* Fix bug in bootstrap (only seen in firefox) */ @@ -78,11 +80,10 @@ dd { /* Section anchors ---------------------------------*/ a.anchor { - margin-left: -30px; - display:inline-block; - width: 30px; - height: 30px; - visibility: hidden; + display: none; + margin-left: 5px; + width: 20px; + height: 20px; background-image: url(./link.svg); background-repeat: no-repeat; @@ -90,17 +91,15 @@ a.anchor { background-position: center center; } -.hasAnchor:hover a.anchor { - visibility: visible; -} - -@media (max-width: 767px) { - .hasAnchor:hover a.anchor { - visibility: hidden; - } +h1:hover .anchor, +h2:hover .anchor, +h3:hover .anchor, +h4:hover .anchor, +h5:hover .anchor, +h6:hover .anchor { + display: inline-block; } - /* Fixes for fixed navbar --------------------------*/ .contents h1, .contents h2, .contents h3, .contents h4 { @@ -264,31 +263,26 @@ table { /* Syntax highlighting ---------------------------------------------------- */ -pre { - word-wrap: normal; - word-break: normal; - border: 1px solid #eee; -} - -pre, code { +pre, code, pre code { background-color: #f8f8f8; color: #333; } +pre, pre code { + white-space: pre-wrap; + word-break: break-all; + overflow-wrap: break-word; +} -pre code { - overflow: auto; - word-wrap: normal; - white-space: pre; +pre { + border: 1px solid #eee; } -pre .img { +pre .img, pre .r-plt { margin: 5px 0; } -pre .img img { +pre .img img, pre .r-plt img { background-color: #fff; - display: block; - height: auto; } code a, pre a { @@ -305,9 +299,8 @@ a.sourceLine:hover { .kw {color: #264D66;} /* keyword */ .co {color: #888888;} /* comment */ -.message { color: black; font-weight: bolder;} -.error { color: orange; font-weight: bolder;} -.warning { color: #6A0366; font-weight: bolder;} +.error {font-weight: bolder;} +.warning {font-weight: bolder;} /* Clipboard --------------------------*/ @@ -365,3 +358,27 @@ mark { content: ""; } } + +/* Section anchors --------------------------------- + Added in pandoc 2.11: https://github.com/jgm/pandoc-templates/commit/9904bf71 +*/ + +div.csl-bib-body { } +div.csl-entry { + clear: both; +} +.hanging-indent div.csl-entry { + margin-left:2em; + text-indent:-2em; +} +div.csl-left-margin { + min-width:2em; + float:left; +} +div.csl-right-inline { + margin-left:2em; + padding-left:1em; +} +div.csl-indent { + margin-left: 2em; +} diff --git a/docs/pkgdown.js b/docs/pkgdown.js index 7e7048fa..6f0eee40 100644 --- a/docs/pkgdown.js +++ b/docs/pkgdown.js @@ -80,7 +80,7 @@ $(document).ready(function() { var copyButton = ""; - $(".examples, div.sourceCode").addClass("hasCopyButton"); + $("div.sourceCode").addClass("hasCopyButton"); // Insert copy buttons: $(copyButton).prependTo(".hasCopyButton"); @@ -91,7 +91,7 @@ // Initialize clipboard: var clipboardBtnCopies = new ClipboardJS('[data-clipboard-copy]', { text: function(trigger) { - return trigger.parentNode.textContent; + return trigger.parentNode.textContent.replace(/\n#>[^\n]*/g, ""); } }); diff --git a/docs/pkgdown.yml b/docs/pkgdown.yml index 44c8b105..e9e64112 100644 --- a/docs/pkgdown.yml +++ b/docs/pkgdown.yml @@ -1,5 +1,5 @@ -pandoc: 2.11.2 -pkgdown: 1.6.1 +pandoc: 2.14.0.3 +pkgdown: 2.0.2 pkgdown_sha: ~ articles: 01-intro: 01-intro.html @@ -14,5 +14,5 @@ articles: 10-group-by: 10-group-by.html 11-custom-group-by: 11-custom-group-by.html 88-trouble-shooting: 88-trouble-shooting.html -last_built: 2021-05-09T09:55Z +last_built: 2022-01-24T10:36Z diff --git a/docs/reference/add_chunk.html b/docs/reference/add_chunk.html index 30f23582..82b43bf9 100644 --- a/docs/reference/add_chunk.html +++ b/docs/reference/add_chunk.html @@ -1,68 +1,13 @@ - - - - - - - -Add a chunk to the disk.frame — add_chunk • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Add a chunk to the disk.frame — add_chunk • disk.frame - - - - + + -
      -
      -
    - - -
    +

    Add a chunk to the disk.frame

    - Source: R/add_chunk.r + Source: R/add_chunk.r
    @@ -167,117 +97,114 @@

    Add a chunk to the disk.frame

    largest numbered file, "n.fst".

    - 
    add_chunk(df, chunk, chunk_id = NULL, full.names = FALSE, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - -
    df

    the disk.frame to add a chunk to

    chunk

    a data.frame to be added as a chunk

    chunk_id

    a numeric number indicating the id of the chunk. If NULL it -will be set to the largest chunk_id + 1

    full.names

    whether the chunk_id name match should be to the full file -path not just the file name

    ...

    Passed in the write_fst. E.g. compress

    - -

    Value

    +
    + 
    add_chunk(df, chunk, chunk_id = NULL, full.names = FALSE, ...)
    +
    +
    +

    Arguments

    +
    df
    +

    the disk.frame to add a chunk to

    +
    chunk
    +

    a data.frame to be added as a chunk

    +
    chunk_id
    +

    a numeric number indicating the id of the chunk. If NULL it +will be set to the largest chunk_id + 1

    +
    full.names
    +

    whether the chunk_id name match should be to the full file +path not just the file name

    +
    ...
    +

    Passed in the write_fst. E.g. compress

    +
    +
    +

    Value

    disk.frame

    -

    Details

    - +
    +
    +

    Details

    The function is the preferred way to add a chunk to a disk.frame. It performs checks on the types to make sure that the new chunk doesn't have different types to the disk.frame.

    +
    -

    Examples

    - 
    # create a disk.frame
-df_path = file.path(tempdir(), "tmp_add_chunk")
-diskf = disk.frame(df_path)
-
-# add a chunk to diskf
-add_chunk(diskf, cars)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP/tmp_add_chunk"
-#> nchunks: 1
-#> nrow (at source): 50
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    add_chunk(diskf, cars)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP/tmp_add_chunk"
-#> nchunks: 2
-#> nrow (at source): 100
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    
-nchunks(diskf) # 2
-
    #> [1] 2
    
-df2 = disk.frame(file.path(tempdir(), "tmp_add_chunk2"))
-
-# add chunks by specifying the chunk_id number; this is especially useful if
-# you wish to add multiple chunk in parralel
-
-add_chunk(df2, data.frame(chunk=1), 1)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP/tmp_add_chunk2"
-#> nchunks: 1
-#> nrow (at source): 1
-#> ncol (at source): 1
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    add_chunk(df2, data.frame(chunk=2), 3)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP/tmp_add_chunk2"
-#> nchunks: 2
-#> nrow (at source): 2
-#> ncol (at source): 1
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    
-nchunks(df2) # 2
-
    #> [1] 2
    
-dir(attr(df2, "path", exact=TRUE))
-
    #> [1] "1.fst" "3.fst"
    # [1] "1.fst" "3.fst"
-
-# clean up
-delete(diskf)
-delete(df2)
-
    +
    +

    Examples

    + 
    # create a disk.frame
+df_path = file.path(tempdir(), "tmp_add_chunk")
+diskf = disk.frame(df_path)
+
+# add a chunk to diskf
+add_chunk(diskf, cars)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5/tmp_add_chunk"
+#> nchunks: 1
+#> nrow (at source): 50
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+add_chunk(diskf, cars)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5/tmp_add_chunk"
+#> nchunks: 2
+#> nrow (at source): 100
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+
+nchunks(diskf) # 2
+#> [1] 2
+
+df2 = disk.frame(file.path(tempdir(), "tmp_add_chunk2"))
+
+# add chunks by specifying the chunk_id number; this is especially useful if
+# you wish to add multiple chunk in parralel
+
+add_chunk(df2, data.frame(chunk=1), 1)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5/tmp_add_chunk2"
+#> nchunks: 1
+#> nrow (at source): 1
+#> ncol (at source): 1
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+add_chunk(df2, data.frame(chunk=2), 3)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5/tmp_add_chunk2"
+#> nchunks: 2
+#> nrow (at source): 2
+#> ncol (at source): 1
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+
+nchunks(df2) # 2
+#> [1] 2
+
+dir(attr(df2, "path", exact=TRUE))
+#> [1] "1.fst" "3.fst"
+# [1] "1.fst" "3.fst"
+
+# clean up
+delete(diskf)
+delete(df2)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/as.data.frame.disk.frame.html b/docs/reference/as.data.frame.disk.frame.html index 275aebaf..abfa6194 100644 --- a/docs/reference/as.data.frame.disk.frame.html +++ b/docs/reference/as.data.frame.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Convert disk.frame to data.frame by collecting all chunks — as.data.frame.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Convert disk.frame to data.frame by collecting all chunks — as.data.frame.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Convert disk.frame to data.frame by collecting all chunks

    - Source: R/as.data.frame.r + Source: R/as.data.frame.r
    @@ -165,113 +95,104 @@

    Convert disk.frame to data.frame by collecting all chunks

    Convert disk.frame to data.frame by collecting all chunks

    - 
    # S3 method for disk.frame
-as.data.frame(x, row.names, optional, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    x

    a disk.frame

    row.names

    NULL or a character vector giving the row names for the data frame. Missing values are not allowed.

    optional

    logical. If TRUE, setting row names and converting column names (to syntactic names: see make.names) is optional. Note that all of R's base package as.data.frame() methods use optional only for column names treatment, basically with the meaning of data.frame(*, check.names = !optional). See also the make.names argument of the matrix method.

    ...

    additional arguments to be passed to or from methods.

    - +
    + 
    # S3 method for disk.frame
+as.data.frame(x, row.names, optional, ...)
    +
    -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-as.data.frame(cars.df)
-
    #>    speed dist
-#> 1      4    2
-#> 2      4   10
-#> 3      7    4
-#> 4      7   22
-#> 5      8   16
-#> 6      9   10
-#> 7     10   18
-#> 8     10   26
-#> 9     10   34
-#> 10    11   17
-#> 11    11   28
-#> 12    12   14
-#> 13    12   20
-#> 14    12   24
-#> 15    12   28
-#> 16    13   26
-#> 17    13   34
-#> 18    13   34
-#> 19    13   46
-#> 20    14   26
-#> 21    14   36
-#> 22    14   60
-#> 23    14   80
-#> 24    15   20
-#> 25    15   26
-#> 26    15   54
-#> 27    16   32
-#> 28    16   40
-#> 29    17   32
-#> 30    17   40
-#> 31    17   50
-#> 32    18   42
-#> 33    18   56
-#> 34    18   76
-#> 35    18   84
-#> 36    19   36
-#> 37    19   46
-#> 38    19   68
-#> 39    20   32
-#> 40    20   48
-#> 41    20   52
-#> 42    20   56
-#> 43    20   64
-#> 44    22   66
-#> 45    23   54
-#> 46    24   70
-#> 47    24   92
-#> 48    24   93
-#> 49    24  120
-#> 50    25   85
    
-# clean up
-delete(cars.df)
-
    +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    row.names
    +

    NULL or a character vector giving the row names for the data frame. Missing values are not allowed.

    +
    optional
    +

    logical. If TRUE, setting row names and converting column names (to syntactic names: see make.names) is optional. Note that all of R's base package as.data.frame() methods use optional only for column names treatment, basically with the meaning of data.frame(*, check.names = !optional). See also the make.names argument of the matrix method.

    +
    ...
    +

    additional arguments to be passed to or from methods.

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+as.data.frame(cars.df)
+#>    speed dist
+#> 1      4    2
+#> 2      4   10
+#> 3      7    4
+#> 4      7   22
+#> 5      8   16
+#> 6      9   10
+#> 7     10   18
+#> 8     10   26
+#> 9     10   34
+#> 10    11   17
+#> 11    11   28
+#> 12    12   14
+#> 13    12   20
+#> 14    12   24
+#> 15    12   28
+#> 16    13   26
+#> 17    13   34
+#> 18    13   34
+#> 19    13   46
+#> 20    14   26
+#> 21    14   36
+#> 22    14   60
+#> 23    14   80
+#> 24    15   20
+#> 25    15   26
+#> 26    15   54
+#> 27    16   32
+#> 28    16   40
+#> 29    17   32
+#> 30    17   40
+#> 31    17   50
+#> 32    18   42
+#> 33    18   56
+#> 34    18   76
+#> 35    18   84
+#> 36    19   36
+#> 37    19   46
+#> 38    19   68
+#> 39    20   32
+#> 40    20   48
+#> 41    20   52
+#> 42    20   56
+#> 43    20   64
+#> 44    22   66
+#> 45    23   54
+#> 46    24   70
+#> 47    24   92
+#> 48    24   93
+#> 49    24  120
+#> 50    25   85
+
+# clean up
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/as.data.table.disk.frame.html b/docs/reference/as.data.table.disk.frame.html index 16530f9e..7e9ac1e2 100644 --- a/docs/reference/as.data.table.disk.frame.html +++ b/docs/reference/as.data.table.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Convert disk.frame to data.table by collecting all chunks — as.data.table.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Convert disk.frame to data.table by collecting all chunks — as.data.table.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Convert disk.frame to data.table by collecting all chunks

    - Source: R/as.data.frame.r + Source: R/as.data.frame.r
    @@ -165,116 +95,112 @@

    Convert disk.frame to data.table by collecting all chunks

    Convert disk.frame to data.table by collecting all chunks

    - 
    # S3 method for disk.frame
-as.data.table(x, keep.rownames = FALSE, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - -
    x

    a disk.frame

    keep.rownames

    passed to as.data.table

    ...

    passed to as.data.table

    - +
    + 
    # S3 method for disk.frame
+as.data.table(x, keep.rownames = FALSE, ...)
    +
    -

    Examples

    - 
    library(data.table)
-
    #> 
-#> Attaching package: 'data.table'
    #> The following object is masked from 'package:purrr':
-#> 
-#>     transpose
    #> The following objects are masked from 'package:dplyr':
-#> 
-#>     between, first, last
    cars.df = as.disk.frame(cars)
-as.data.table(cars.df)
-
    #>     speed dist
-#>  1:     4    2
-#>  2:     4   10
-#>  3:     7    4
-#>  4:     7   22
-#>  5:     8   16
-#>  6:     9   10
-#>  7:    10   18
-#>  8:    10   26
-#>  9:    10   34
-#> 10:    11   17
-#> 11:    11   28
-#> 12:    12   14
-#> 13:    12   20
-#> 14:    12   24
-#> 15:    12   28
-#> 16:    13   26
-#> 17:    13   34
-#> 18:    13   34
-#> 19:    13   46
-#> 20:    14   26
-#> 21:    14   36
-#> 22:    14   60
-#> 23:    14   80
-#> 24:    15   20
-#> 25:    15   26
-#> 26:    15   54
-#> 27:    16   32
-#> 28:    16   40
-#> 29:    17   32
-#> 30:    17   40
-#> 31:    17   50
-#> 32:    18   42
-#> 33:    18   56
-#> 34:    18   76
-#> 35:    18   84
-#> 36:    19   36
-#> 37:    19   46
-#> 38:    19   68
-#> 39:    20   32
-#> 40:    20   48
-#> 41:    20   52
-#> 42:    20   56
-#> 43:    20   64
-#> 44:    22   66
-#> 45:    23   54
-#> 46:    24   70
-#> 47:    24   92
-#> 48:    24   93
-#> 49:    24  120
-#> 50:    25   85
-#>     speed dist
    
-# clean up
-delete(cars.df)
-
    +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    keep.rownames
    +

    passed to as.data.table

    +
    ...
    +

    passed to as.data.table

    +
    + +
    +

    Examples

    + 
    library(data.table)
+#> 
+#> Attaching package: 'data.table'
+#> The following object is masked from 'package:purrr':
+#> 
+#>     transpose
+#> The following objects are masked from 'package:dplyr':
+#> 
+#>     between, first, last
+cars.df = as.disk.frame(cars)
+as.data.table(cars.df)
+#>     speed dist
+#>  1:     4    2
+#>  2:     4   10
+#>  3:     7    4
+#>  4:     7   22
+#>  5:     8   16
+#>  6:     9   10
+#>  7:    10   18
+#>  8:    10   26
+#>  9:    10   34
+#> 10:    11   17
+#> 11:    11   28
+#> 12:    12   14
+#> 13:    12   20
+#> 14:    12   24
+#> 15:    12   28
+#> 16:    13   26
+#> 17:    13   34
+#> 18:    13   34
+#> 19:    13   46
+#> 20:    14   26
+#> 21:    14   36
+#> 22:    14   60
+#> 23:    14   80
+#> 24:    15   20
+#> 25:    15   26
+#> 26:    15   54
+#> 27:    16   32
+#> 28:    16   40
+#> 29:    17   32
+#> 30:    17   40
+#> 31:    17   50
+#> 32:    18   42
+#> 33:    18   56
+#> 34:    18   76
+#> 35:    18   84
+#> 36:    19   36
+#> 37:    19   46
+#> 38:    19   68
+#> 39:    20   32
+#> 40:    20   48
+#> 41:    20   52
+#> 42:    20   56
+#> 43:    20   64
+#> 44:    22   66
+#> 45:    23   54
+#> 46:    24   70
+#> 47:    24   92
+#> 48:    24   93
+#> 49:    24  120
+#> 50:    25   85
+#>     speed dist
+
+# clean up
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/as.disk.frame.html b/docs/reference/as.disk.frame.html index b6f779b8..10189939 100644 --- a/docs/reference/as.disk.frame.html +++ b/docs/reference/as.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Make a data.frame into a disk.frame — as.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Make a data.frame into a disk.frame — as.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Make a data.frame into a disk.frame

    - Source: R/as.disk.frame.r + Source: R/as.disk.frame.r
    @@ -165,91 +95,75 @@

    Make a data.frame into a disk.frame

    Make a data.frame into a disk.frame

    - 
    as.disk.frame(
-  df,
-  outdir = tempfile(fileext = ".df"),
-  nchunks = recommend_nchunks(df),
-  overwrite = FALSE,
-  shardby = NULL,
-  compress = 50,
-  ...
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df

    a disk.frame

    outdir

    the output directory

    nchunks

    number of chunks

    overwrite

    if TRUE the outdir will be overwritten, if FALSE it will throw an error if the directory is not empty

    shardby

    The shardkey

    compress

    the compression level 0-100; 100 is highest

    ...

    passed to output_disk.frame

    - - -

    Examples

    - 
    # write to temporary location
-cars.df = as.disk.frame(cars) 
-
-# specify a different path in the temporary folder, you are free to choose a different folder
-cars_new_location.df = as.disk.frame(cars, outdir = file.path(tempdir(), "some_path.df"))
-
-# specify a different number of chunks
-# this writes to tempdir() by default
-cars_chunks.df = as.disk.frame(cars, nchunks = 4, overwrite = TRUE) 
-
-# clean up
-delete(cars.df)
-delete(cars_new_location.df)
-delete(cars_chunks.df)
-
    +
    + 
    as.disk.frame(
+  df,
+  outdir = tempfile(fileext = ".df"),
+  nchunks = recommend_nchunks(df),
+  overwrite = FALSE,
+  shardby = NULL,
+  compress = 50,
+  ...
+)
    +
    + +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    outdir
    +

    the output directory

    +
    nchunks
    +

    number of chunks

    +
    overwrite
    +

    if TRUE the outdir will be overwritten, if FALSE it will throw an error if the directory is not empty

    +
    shardby
    +

    The shardkey

    +
    compress
    +

    the compression level 0-100; 100 is highest

    +
    ...
    +

    passed to output_disk.frame

    +
    + +
    +

    Examples

    + 
    # write to temporary location
+cars.df = as.disk.frame(cars) 
+
+# specify a different path in the temporary folder, you are free to choose a different folder
+cars_new_location.df = as.disk.frame(cars, outdir = file.path(tempdir(), "some_path.df"))
+
+# specify a different number of chunks
+# this writes to tempdir() by default
+cars_chunks.df = as.disk.frame(cars, nchunks = 4, overwrite = TRUE) 
+
+# clean up
+delete(cars.df)
+delete(cars_new_location.df)
+delete(cars_chunks.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/chunk_group_by.html b/docs/reference/chunk_group_by.html index 731d94e4..3b4f8168 100644 --- a/docs/reference/chunk_group_by.html +++ b/docs/reference/chunk_group_by.html @@ -1,70 +1,15 @@ - - - - - - - -Group by within each disk.frame — chunk_summarize • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Group by within each disk.frame — chunk_summarize • disk.frame - - - - - - - - - - - + + - - -
    -
    - -
    - -
    +

    Group by within each disk.frame

    - Source: R/dplyr_verbs.r + Source: R/dplyr_verbs.r
    @@ -171,56 +101,49 @@

    Group by within each disk.frame

    reorganizes the chunks by the shard key.

    - 
    chunk_summarize(.data, ...)
-
-chunk_summarise(.data, ...)
-
-chunk_group_by(.data, ...)
+    
    
+    
    chunk_summarize(.data, ...)
 
-chunk_ungroup(.data, ...)
    
+chunk_summarise(.data, ...)
 
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
    .data
    a disk.frame
    ...
    passed to dplyr::group_by
    
+chunk_group_by(.data, ...)
 
-    
    See also
    
+chunk_ungroup(.data, ...)
    +
    -

    hard_group_by group_by

    +
    +

    Arguments

    +
    .data
    +

    a disk.frame

    +
    ...
    +

    passed to dplyr::group_by

    +
    +
    +

    See also

    +

    hard_group_by group_by

    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/cmap.html b/docs/reference/cmap.html index c2cd39fa..8d46d751 100644 --- a/docs/reference/cmap.html +++ b/docs/reference/cmap.html @@ -1,71 +1,16 @@ - - - - - - - -Apply the same function to all chunks — cmap • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Apply the same function to all chunks — cmap • disk.frame - - - - - - - - - - - + + - - -
    -
    - -
    - -
    +

    Apply the same function to all chunks

    - Source: R/cmap.r, R/map-deprecated.r + Source: R/cmap.r, R/map-deprecated.r
    @@ -173,439 +103,415 @@

    Apply the same function to all chunks

    `delayed` is an alias for lazy and is consistent with the naming in Dask and Dagger.jl

    - 
    cmap(.x, .f, ...)
-
-# S3 method for disk.frame
-cmap(
-  .x,
-  .f,
-  ...,
-  outdir = NULL,
-  keep = NULL,
-  chunks = nchunks(.x),
-  compress = 50,
-  lazy = TRUE,
-  overwrite = FALSE,
-  vars_and_pkgs = future::getGlobalsAndPackages(.f, envir = parent.frame()),
-  .progress = TRUE
-)
-
-cmap_dfr(.x, .f, ..., .id = NULL)
-
-# S3 method for disk.frame
-cmap_dfr(.x, .f, ..., .id = NULL, use.names = fill, fill = FALSE, idcol = NULL)
-
-cimap(.x, .f, ...)
-
-# S3 method for disk.frame
-cimap(
-  .x,
-  .f,
-  outdir = NULL,
-  keep = NULL,
-  chunks = nchunks(.x),
-  compress = 50,
-  lazy = TRUE,
-  overwrite = FALSE,
-  ...
-)
-
-cimap_dfr(.x, .f, ..., .id = NULL)
-
-# S3 method for disk.frame
-cimap_dfr(
-  .x,
-  .f,
-  ...,
-  .id = NULL,
-  use.names = fill,
-  fill = FALSE,
-  idcol = NULL
-)
-
-lazy(.x, .f, ...)
-
-# S3 method for disk.frame
-lazy(.x, .f, ...)
-
-delayed(.x, .f, ...)
-
-chunk_lapply(...)
-
-map(.x, .f, ...)
-
-# S3 method for disk.frame
-map(...)
-
-# S3 method for default
-map(.x, .f, ...)
-
-imap_dfr(.x, .f, ..., .id = NULL)
-
-# S3 method for disk.frame
-imap_dfr(...)
-
-# S3 method for default
-imap_dfr(.x, .f, ..., .id = NULL)
-
-imap(.x, .f, ...)
-
-# S3 method for default
-imap(.x, .f, ...)
-
-# S3 method for disk.frame
-map_dfr(...)
-
-# S3 method for default
-map_dfr(.x, .f, ..., .id = NULL)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    .x

    a disk.frame

    .f

    a function to apply to each of the chunks

    ...

    for compatibility with `purrr::map`

    outdir

    the output directory

    keep

    the columns to keep from the input

    chunks

    The number of chunks to output

    compress

    0-100 fst compression ratio

    lazy

    if TRUE then do this lazily

    overwrite

    if TRUE removes any existing chunks in the data

    vars_and_pkgs

    variables and packages to send to a background session. This is typically automatically detected

    .progress

    A logical, for whether or not to print a progress bar for multiprocess, multisession, and multicore plans. From furrr

    .id

    not used

    use.names

    for cmap_dfr's call to data.table::rbindlist. See data.table::rbindlist

    fill

    for cmap_dfr's call to data.table::rbindlist. See data.table::rbindlist

    idcol

    for cmap_dfr's call to data.table::rbindlist. See data.table::rbindlist

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-
-# return the first row of each chunk lazily
-# 
-cars2 = cmap(cars.df, function(chunk) {
- chunk[,1]
-})
-
-collect(cars2)
-
    #>     speed
-#>  1:     4
-#>  2:     4
-#>  3:     7
-#>  4:     7
-#>  5:     8
-#>  6:     9
-#>  7:    10
-#>  8:    10
-#>  9:    10
-#> 10:    11
-#> 11:    11
-#> 12:    12
-#> 13:    12
-#> 14:    12
-#> 15:    12
-#> 16:    13
-#> 17:    13
-#> 18:    13
-#> 19:    13
-#> 20:    14
-#> 21:    14
-#> 22:    14
-#> 23:    14
-#> 24:    15
-#> 25:    15
-#> 26:    15
-#> 27:    16
-#> 28:    16
-#> 29:    17
-#> 30:    17
-#> 31:    17
-#> 32:    18
-#> 33:    18
-#> 34:    18
-#> 35:    18
-#> 36:    19
-#> 37:    19
-#> 38:    19
-#> 39:    20
-#> 40:    20
-#> 41:    20
-#> 42:    20
-#> 43:    20
-#> 44:    22
-#> 45:    23
-#> 46:    24
-#> 47:    24
-#> 48:    24
-#> 49:    24
-#> 50:    25
-#>     speed
    
-# same as above but using purrr 
-cars2 = cmap(cars.df, ~.x[1,])
-
-collect(cars2)
-
    #>    speed dist
-#> 1:     4    2
-#> 2:    11   17
-#> 3:    13   46
-#> 4:    16   40
-#> 5:    19   46
-#> 6:    24   70
    
-# return the first row of each chunk eagerly as list
-cmap(cars.df, ~.x[1,], lazy = FALSE)
-
    #> [[1]]
-#>    speed dist
-#> 1:     4    2
-#> 
-#> [[2]]
-#>    speed dist
-#> 1:    11   17
-#> 
-#> [[3]]
-#>    speed dist
-#> 1:    13   46
-#> 
-#> [[4]]
-#>    speed dist
-#> 1:    16   40
-#> 
-#> [[5]]
-#>    speed dist
-#> 1:    19   46
-#> 
-#> [[6]]
-#>    speed dist
-#> 1:    24   70
-#> 
    
-# return the first row of each chunk eagerly as data.table/data.frame by row-binding
-cmap_dfr(cars.df, ~.x[1,])
-
    #>    speed dist
-#> 1:     4    2
-#> 2:    11   17
-#> 3:    13   46
-#> 4:    16   40
-#> 5:    19   46
-#> 6:    24   70
    
-# lazy and delayed are just an aliases for cmap(..., lazy = TRUE)
-collect(lazy(cars.df, ~.x[1,]))
-
    #>    speed dist
-#> 1:     4    2
-#> 2:    11   17
-#> 3:    13   46
-#> 4:    16   40
-#> 5:    19   46
-#> 6:    24   70
    collect(delayed(cars.df, ~.x[1,]))
-
    #>    speed dist
-#> 1:     4    2
-#> 2:    11   17
-#> 3:    13   46
-#> 4:    16   40
-#> 5:    19   46
-#> 6:    24   70
    
-# clean up cars.df
-delete(cars.df)
-cars.df = as.disk.frame(cars)
-
-# .x is the chunk and .y is the ID as an integer
-
-# lazy = TRUE support is not available at the moment
-cimap(cars.df, ~.x[, id := .y], lazy = FALSE)
-
    #> [[1]]
-#>    speed dist id
-#> 1:     4    2  1
-#> 2:     4   10  1
-#> 3:     7    4  1
-#> 4:     7   22  1
-#> 5:     8   16  1
-#> 6:     9   10  1
-#> 7:    10   18  1
-#> 8:    10   26  1
-#> 9:    10   34  1
-#> 
-#> [[2]]
-#>    speed dist id
-#> 1:    11   17  2
-#> 2:    11   28  2
-#> 3:    12   14  2
-#> 4:    12   20  2
-#> 5:    12   24  2
-#> 6:    12   28  2
-#> 7:    13   26  2
-#> 8:    13   34  2
-#> 9:    13   34  2
-#> 
-#> [[3]]
-#>    speed dist id
-#> 1:    13   46  3
-#> 2:    14   26  3
-#> 3:    14   36  3
-#> 4:    14   60  3
-#> 5:    14   80  3
-#> 6:    15   20  3
-#> 7:    15   26  3
-#> 8:    15   54  3
-#> 9:    16   32  3
-#> 
-#> [[4]]
-#>    speed dist id
-#> 1:    16   40  4
-#> 2:    17   32  4
-#> 3:    17   40  4
-#> 4:    17   50  4
-#> 5:    18   42  4
-#> 6:    18   56  4
-#> 7:    18   76  4
-#> 8:    18   84  4
-#> 9:    19   36  4
-#> 
-#> [[5]]
-#>    speed dist id
-#> 1:    19   46  5
-#> 2:    19   68  5
-#> 3:    20   32  5
-#> 4:    20   48  5
-#> 5:    20   52  5
-#> 6:    20   56  5
-#> 7:    20   64  5
-#> 8:    22   66  5
-#> 9:    23   54  5
-#> 
-#> [[6]]
-#>    speed dist id
-#> 1:    24   70  6
-#> 2:    24   92  6
-#> 3:    24   93  6
-#> 4:    24  120  6
-#> 5:    25   85  6
-#> 
    
-cimap_dfr(cars.df, ~.x[, id := .y])
-
    #>     speed dist id
-#>  1:     4    2  1
-#>  2:     4   10  1
-#>  3:     7    4  1
-#>  4:     7   22  1
-#>  5:     8   16  1
-#>  6:     9   10  1
-#>  7:    10   18  1
-#>  8:    10   26  1
-#>  9:    10   34  1
-#> 10:    11   17  2
-#> 11:    11   28  2
-#> 12:    12   14  2
-#> 13:    12   20  2
-#> 14:    12   24  2
-#> 15:    12   28  2
-#> 16:    13   26  2
-#> 17:    13   34  2
-#> 18:    13   34  2
-#> 19:    13   46  3
-#> 20:    14   26  3
-#> 21:    14   36  3
-#> 22:    14   60  3
-#> 23:    14   80  3
-#> 24:    15   20  3
-#> 25:    15   26  3
-#> 26:    15   54  3
-#> 27:    16   32  3
-#> 28:    16   40  4
-#> 29:    17   32  4
-#> 30:    17   40  4
-#> 31:    17   50  4
-#> 32:    18   42  4
-#> 33:    18   56  4
-#> 34:    18   76  4
-#> 35:    18   84  4
-#> 36:    19   36  4
-#> 37:    19   46  5
-#> 38:    19   68  5
-#> 39:    20   32  5
-#> 40:    20   48  5
-#> 41:    20   52  5
-#> 42:    20   56  5
-#> 43:    20   64  5
-#> 44:    22   66  5
-#> 45:    23   54  5
-#> 46:    24   70  6
-#> 47:    24   92  6
-#> 48:    24   93  6
-#> 49:    24  120  6
-#> 50:    25   85  6
-#>     speed dist id
    
-# clean up cars.df
-delete(cars.df)
-
    +
    + 
    cmap(.x, .f, ...)
+
+# S3 method for disk.frame
+cmap(
+  .x,
+  .f,
+  ...,
+  outdir = NULL,
+  keep = NULL,
+  chunks = nchunks(.x),
+  compress = 50,
+  lazy = TRUE,
+  overwrite = FALSE,
+  vars_and_pkgs = future::getGlobalsAndPackages(.f, envir = parent.frame()),
+  .progress = TRUE
+)
+
+cmap_dfr(.x, .f, ..., .id = NULL)
+
+# S3 method for disk.frame
+cmap_dfr(.x, .f, ..., .id = NULL, use.names = fill, fill = FALSE, idcol = NULL)
+
+cimap(.x, .f, ...)
+
+# S3 method for disk.frame
+cimap(
+  .x,
+  .f,
+  outdir = NULL,
+  keep = NULL,
+  chunks = nchunks(.x),
+  compress = 50,
+  lazy = TRUE,
+  overwrite = FALSE,
+  ...
+)
+
+cimap_dfr(.x, .f, ..., .id = NULL)
+
+# S3 method for disk.frame
+cimap_dfr(
+  .x,
+  .f,
+  ...,
+  .id = NULL,
+  use.names = fill,
+  fill = FALSE,
+  idcol = NULL
+)
+
+lazy(.x, .f, ...)
+
+# S3 method for disk.frame
+lazy(.x, .f, ...)
+
+delayed(.x, .f, ...)
+
+chunk_lapply(...)
+
+map(.x, .f, ...)
+
+# S3 method for disk.frame
+map(...)
+
+# S3 method for default
+map(.x, .f, ...)
+
+imap_dfr(.x, .f, ..., .id = NULL)
+
+# S3 method for disk.frame
+imap_dfr(...)
+
+# S3 method for default
+imap_dfr(.x, .f, ..., .id = NULL)
+
+imap(.x, .f, ...)
+
+# S3 method for default
+imap(.x, .f, ...)
+
+# S3 method for disk.frame
+map_dfr(...)
+
+# S3 method for default
+map_dfr(.x, .f, ..., .id = NULL)
    +
    + +
    +

    Arguments

    +
    .x
    +

    a disk.frame

    +
    .f
    +

    a function to apply to each of the chunks

    +
    ...
    +

    for compatibility with `purrr::map`

    +
    outdir
    +

    the output directory

    +
    keep
    +

    the columns to keep from the input

    +
    chunks
    +

    The number of chunks to output

    +
    compress
    +

    0-100 fst compression ratio

    +
    lazy
    +

    if TRUE then do this lazily

    +
    overwrite
    +

    if TRUE removes any existing chunks in the data

    +
    vars_and_pkgs
    +

    variables and packages to send to a background session. This is typically automatically detected

    +
    .progress
    +

    A logical, for whether or not to print a progress bar for multiprocess, multisession, and multicore plans. From furrr

    +
    .id
    +

    not used

    +
    use.names
    +

    for cmap_dfr's call to data.table::rbindlist. See data.table::rbindlist

    +
    fill
    +

    for cmap_dfr's call to data.table::rbindlist. See data.table::rbindlist

    +
    idcol
    +

    for cmap_dfr's call to data.table::rbindlist. See data.table::rbindlist

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# return the first row of each chunk lazily
+# 
+cars2 = cmap(cars.df, function(chunk) {
+ chunk[,1]
+})
+
+collect(cars2)
+#>     speed
+#>  1:     4
+#>  2:     4
+#>  3:     7
+#>  4:     7
+#>  5:     8
+#>  6:     9
+#>  7:    10
+#>  8:    10
+#>  9:    10
+#> 10:    11
+#> 11:    11
+#> 12:    12
+#> 13:    12
+#> 14:    12
+#> 15:    12
+#> 16:    13
+#> 17:    13
+#> 18:    13
+#> 19:    13
+#> 20:    14
+#> 21:    14
+#> 22:    14
+#> 23:    14
+#> 24:    15
+#> 25:    15
+#> 26:    15
+#> 27:    16
+#> 28:    16
+#> 29:    17
+#> 30:    17
+#> 31:    17
+#> 32:    18
+#> 33:    18
+#> 34:    18
+#> 35:    18
+#> 36:    19
+#> 37:    19
+#> 38:    19
+#> 39:    20
+#> 40:    20
+#> 41:    20
+#> 42:    20
+#> 43:    20
+#> 44:    22
+#> 45:    23
+#> 46:    24
+#> 47:    24
+#> 48:    24
+#> 49:    24
+#> 50:    25
+#>     speed
+
+# same as above but using purrr 
+cars2 = cmap(cars.df, ~.x[1,])
+
+collect(cars2)
+#>    speed dist
+#> 1:     4    2
+#> 2:    11   17
+#> 3:    13   46
+#> 4:    16   40
+#> 5:    19   46
+#> 6:    24   70
+
+# return the first row of each chunk eagerly as list
+cmap(cars.df, ~.x[1,], lazy = FALSE)
+#> [[1]]
+#>    speed dist
+#> 1:     4    2
+#> 
+#> [[2]]
+#>    speed dist
+#> 1:    11   17
+#> 
+#> [[3]]
+#>    speed dist
+#> 1:    13   46
+#> 
+#> [[4]]
+#>    speed dist
+#> 1:    16   40
+#> 
+#> [[5]]
+#>    speed dist
+#> 1:    19   46
+#> 
+#> [[6]]
+#>    speed dist
+#> 1:    24   70
+#> 
+
+# return the first row of each chunk eagerly as data.table/data.frame by row-binding
+cmap_dfr(cars.df, ~.x[1,])
+#>    speed dist
+#> 1:     4    2
+#> 2:    11   17
+#> 3:    13   46
+#> 4:    16   40
+#> 5:    19   46
+#> 6:    24   70
+
+# lazy and delayed are just an aliases for cmap(..., lazy = TRUE)
+collect(lazy(cars.df, ~.x[1,]))
+#>    speed dist
+#> 1:     4    2
+#> 2:    11   17
+#> 3:    13   46
+#> 4:    16   40
+#> 5:    19   46
+#> 6:    24   70
+collect(delayed(cars.df, ~.x[1,]))
+#>    speed dist
+#> 1:     4    2
+#> 2:    11   17
+#> 3:    13   46
+#> 4:    16   40
+#> 5:    19   46
+#> 6:    24   70
+
+# clean up cars.df
+delete(cars.df)
+cars.df = as.disk.frame(cars)
+
+# .x is the chunk and .y is the ID as an integer
+
+# lazy = TRUE support is not available at the moment
+cimap(cars.df, ~.x[, id := .y], lazy = FALSE)
+#> [[1]]
+#>    speed dist id
+#> 1:     4    2  1
+#> 2:     4   10  1
+#> 3:     7    4  1
+#> 4:     7   22  1
+#> 5:     8   16  1
+#> 6:     9   10  1
+#> 7:    10   18  1
+#> 8:    10   26  1
+#> 9:    10   34  1
+#> 
+#> [[2]]
+#>    speed dist id
+#> 1:    11   17  2
+#> 2:    11   28  2
+#> 3:    12   14  2
+#> 4:    12   20  2
+#> 5:    12   24  2
+#> 6:    12   28  2
+#> 7:    13   26  2
+#> 8:    13   34  2
+#> 9:    13   34  2
+#> 
+#> [[3]]
+#>    speed dist id
+#> 1:    13   46  3
+#> 2:    14   26  3
+#> 3:    14   36  3
+#> 4:    14   60  3
+#> 5:    14   80  3
+#> 6:    15   20  3
+#> 7:    15   26  3
+#> 8:    15   54  3
+#> 9:    16   32  3
+#> 
+#> [[4]]
+#>    speed dist id
+#> 1:    16   40  4
+#> 2:    17   32  4
+#> 3:    17   40  4
+#> 4:    17   50  4
+#> 5:    18   42  4
+#> 6:    18   56  4
+#> 7:    18   76  4
+#> 8:    18   84  4
+#> 9:    19   36  4
+#> 
+#> [[5]]
+#>    speed dist id
+#> 1:    19   46  5
+#> 2:    19   68  5
+#> 3:    20   32  5
+#> 4:    20   48  5
+#> 5:    20   52  5
+#> 6:    20   56  5
+#> 7:    20   64  5
+#> 8:    22   66  5
+#> 9:    23   54  5
+#> 
+#> [[6]]
+#>    speed dist id
+#> 1:    24   70  6
+#> 2:    24   92  6
+#> 3:    24   93  6
+#> 4:    24  120  6
+#> 5:    25   85  6
+#> 
+
+cimap_dfr(cars.df, ~.x[, id := .y])
+#>     speed dist id
+#>  1:     4    2  1
+#>  2:     4   10  1
+#>  3:     7    4  1
+#>  4:     7   22  1
+#>  5:     8   16  1
+#>  6:     9   10  1
+#>  7:    10   18  1
+#>  8:    10   26  1
+#>  9:    10   34  1
+#> 10:    11   17  2
+#> 11:    11   28  2
+#> 12:    12   14  2
+#> 13:    12   20  2
+#> 14:    12   24  2
+#> 15:    12   28  2
+#> 16:    13   26  2
+#> 17:    13   34  2
+#> 18:    13   34  2
+#> 19:    13   46  3
+#> 20:    14   26  3
+#> 21:    14   36  3
+#> 22:    14   60  3
+#> 23:    14   80  3
+#> 24:    15   20  3
+#> 25:    15   26  3
+#> 26:    15   54  3
+#> 27:    16   32  3
+#> 28:    16   40  4
+#> 29:    17   32  4
+#> 30:    17   40  4
+#> 31:    17   50  4
+#> 32:    18   42  4
+#> 33:    18   56  4
+#> 34:    18   76  4
+#> 35:    18   84  4
+#> 36:    19   36  4
+#> 37:    19   46  5
+#> 38:    19   68  5
+#> 39:    20   32  5
+#> 40:    20   48  5
+#> 41:    20   52  5
+#> 42:    20   56  5
+#> 43:    20   64  5
+#> 44:    22   66  5
+#> 45:    23   54  5
+#> 46:    24   70  6
+#> 47:    24   92  6
+#> 48:    24   93  6
+#> 49:    24  120  6
+#> 50:    25   85  6
+#>     speed dist id
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/cmap2.html b/docs/reference/cmap2.html index 35f65ab5..0910f609 100644 --- a/docs/reference/cmap2.html +++ b/docs/reference/cmap2.html @@ -1,68 +1,13 @@ - - - - - - - -`cmap2` a function to two disk.frames — cmap2 • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -`cmap2` a function to two disk.frames — cmap2 • disk.frame + + - - - - -
    -
    - -
    - -
    +

    `cmap2` a function to two disk.frames

    - Source: R/map2.r, R/map_by_chunk_id.r + Source: R/map2.r, R/map_by_chunk_id.r
    @@ -167,175 +97,164 @@

    `cmap2` a function to two disk.frames

    gets run by .f(x.chunk, y.chunk)

    - 
    cmap2(.x, .y, .f, ...)
-
-map2(.x, .y, .f, ...)
-
-map_by_chunk_id(.x, .y, .f, ..., outdir)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - -
    .x

    a disk.frame

    .y

    a disk.frame

    .f

    a function to be called on each chunk of x and y matched by -chunk_id

    ...

    not used

    outdir

    output directory

    +
    + 
    cmap2(.x, .y, .f, ...)
 
+map2(.x, .y, .f, ...)
 
-    
    Examples
    
-    
    cars.df = as.disk.frame(cars)
+map_by_chunk_id(.x, .y, .f, ..., outdir)
    +
    -cars2.df = cmap2(cars.df, cars.df, ~data.table::rbindlist(list(.x, .y))) -collect(cars2.df) -
    #> speed dist -#> 1: 4 2 -#> 2: 4 10 -#> 3: 7 4 -#> 4: 7 22 -#> 5: 8 16 -#> 6: 9 10 -#> 7: 10 18 -#> 8: 10 26 -#> 9: 10 34 -#> 10: 4 2 -#> 11: 4 10 -#> 12: 7 4 -#> 13: 7 22 -#> 14: 8 16 -#> 15: 9 10 -#> 16: 10 18 -#> 17: 10 26 -#> 18: 10 34 -#> 19: 11 17 -#> 20: 11 28 -#> 21: 12 14 -#> 22: 12 20 -#> 23: 12 24 -#> 24: 12 28 -#> 25: 13 26 -#> 26: 13 34 -#> 27: 13 34 -#> 28: 11 17 -#> 29: 11 28 -#> 30: 12 14 -#> 31: 12 20 -#> 32: 12 24 -#> 33: 12 28 -#> 34: 13 26 -#> 35: 13 34 -#> 36: 13 34 -#> 37: 13 46 -#> 38: 14 26 -#> 39: 14 36 -#> 40: 14 60 -#> 41: 14 80 -#> 42: 15 20 -#> 43: 15 26 -#> 44: 15 54 -#> 45: 16 32 -#> 46: 13 46 -#> 47: 14 26 -#> 48: 14 36 -#> 49: 14 60 -#> 50: 14 80 -#> 51: 15 20 -#> 52: 15 26 -#> 53: 15 54 -#> 54: 16 32 -#> 55: 16 40 -#> 56: 17 32 -#> 57: 17 40 -#> 58: 17 50 -#> 59: 18 42 -#> 60: 18 56 -#> 61: 18 76 -#> 62: 18 84 -#> 63: 19 36 -#> 64: 16 40 -#> 65: 17 32 -#> 66: 17 40 -#> 67: 17 50 -#> 68: 18 42 -#> 69: 18 56 -#> 70: 18 76 -#> 71: 18 84 -#> 72: 19 36 -#> 73: 19 46 -#> 74: 19 68 -#> 75: 20 32 -#> 76: 20 48 -#> 77: 20 52 -#> 78: 20 56 -#> 79: 20 64 -#> 80: 22 66 -#> 81: 23 54 -#> 82: 19 46 -#> 83: 19 68 -#> 84: 20 32 -#> 85: 20 48 -#> 86: 20 52 -#> 87: 20 56 -#> 88: 20 64 -#> 89: 22 66 -#> 90: 23 54 -#> 91: 24 70 -#> 92: 24 92 -#> 93: 24 93 -#> 94: 24 120 -#> 95: 25 85 -#> 96: 24 70 -#> 97: 24 92 -#> 98: 24 93 -#> 99: 24 120 -#> 100: 25 85 -#> speed dist
    -# clean up cars.df -delete(cars.df) -delete(cars2.df) -
    +
    +

    Arguments

    +
    .x
    +

    a disk.frame

    +
    .y
    +

    a disk.frame

    +
    .f
    +

    a function to be called on each chunk of x and y matched by +chunk_id

    +
    ...
    +

    not used

    +
    outdir
    +

    output directory

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+cars2.df = cmap2(cars.df, cars.df, ~data.table::rbindlist(list(.x, .y)))
+collect(cars2.df)
+#>      speed dist
+#>   1:     4    2
+#>   2:     4   10
+#>   3:     7    4
+#>   4:     7   22
+#>   5:     8   16
+#>   6:     9   10
+#>   7:    10   18
+#>   8:    10   26
+#>   9:    10   34
+#>  10:     4    2
+#>  11:     4   10
+#>  12:     7    4
+#>  13:     7   22
+#>  14:     8   16
+#>  15:     9   10
+#>  16:    10   18
+#>  17:    10   26
+#>  18:    10   34
+#>  19:    11   17
+#>  20:    11   28
+#>  21:    12   14
+#>  22:    12   20
+#>  23:    12   24
+#>  24:    12   28
+#>  25:    13   26
+#>  26:    13   34
+#>  27:    13   34
+#>  28:    11   17
+#>  29:    11   28
+#>  30:    12   14
+#>  31:    12   20
+#>  32:    12   24
+#>  33:    12   28
+#>  34:    13   26
+#>  35:    13   34
+#>  36:    13   34
+#>  37:    13   46
+#>  38:    14   26
+#>  39:    14   36
+#>  40:    14   60
+#>  41:    14   80
+#>  42:    15   20
+#>  43:    15   26
+#>  44:    15   54
+#>  45:    16   32
+#>  46:    13   46
+#>  47:    14   26
+#>  48:    14   36
+#>  49:    14   60
+#>  50:    14   80
+#>  51:    15   20
+#>  52:    15   26
+#>  53:    15   54
+#>  54:    16   32
+#>  55:    16   40
+#>  56:    17   32
+#>  57:    17   40
+#>  58:    17   50
+#>  59:    18   42
+#>  60:    18   56
+#>  61:    18   76
+#>  62:    18   84
+#>  63:    19   36
+#>  64:    16   40
+#>  65:    17   32
+#>  66:    17   40
+#>  67:    17   50
+#>  68:    18   42
+#>  69:    18   56
+#>  70:    18   76
+#>  71:    18   84
+#>  72:    19   36
+#>  73:    19   46
+#>  74:    19   68
+#>  75:    20   32
+#>  76:    20   48
+#>  77:    20   52
+#>  78:    20   56
+#>  79:    20   64
+#>  80:    22   66
+#>  81:    23   54
+#>  82:    19   46
+#>  83:    19   68
+#>  84:    20   32
+#>  85:    20   48
+#>  86:    20   52
+#>  87:    20   56
+#>  88:    20   64
+#>  89:    22   66
+#>  90:    23   54
+#>  91:    24   70
+#>  92:    24   92
+#>  93:    24   93
+#>  94:    24  120
+#>  95:    25   85
+#>  96:    24   70
+#>  97:    24   92
+#>  98:    24   93
+#>  99:    24  120
+#> 100:    25   85
+#>      speed dist
+
+# clean up cars.df
+delete(cars.df)
+delete(cars2.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/collect.html b/docs/reference/collect.html index 01995d45..5b195078 100644 --- a/docs/reference/collect.html +++ b/docs/reference/collect.html @@ -1,70 +1,15 @@ - - - - - - - -Bring the disk.frame into R — collect.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Bring the disk.frame into R — collect.disk.frame • disk.frame - - - - - - - - - + + - - - - -
    -
    - -
    - -
    +

    Bring the disk.frame into R

    - Source: R/collect.r, R/collect.summarized_disk.frame.r + Source: R/collect.r, R/collect.summarized_disk.frame.r
    @@ -171,211 +101,205 @@

    Bring the disk.frame into R

    operations as data.table/data.frame or as a list

    - 
    # S3 method for disk.frame
-collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
+    
    
+    
    # S3 method for disk.frame
+collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
 
-collect_list(x, simplify = FALSE, parallel = !is.null(attr(x, "lazyfn")))
+collect_list(x, simplify = FALSE, parallel = !is.null(attr(x, "lazyfn")))
 
-# S3 method for summarized_disk.frame
-collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
    
+# S3 method for summarized_disk.frame
+collect(x, ..., parallel = !is.null(attr(x, "lazyfn")))
    +
    -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    x

    a disk.frame

    ...

    not used

    parallel

    if TRUE the collection is performed in parallel. By default +

    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    ...
    +

    not used

    +
    parallel
    +

    if TRUE the collection is performed in parallel. By default if there are delayed/lazy steps then it will be parallel, otherwise it will not be in parallel. This is because parallel requires transferring data from background R session to the current R session and if there is no computation then it's better to avoid transferring data between session, -hence parallel = FALSE is a better choice

    simplify

    Should the result be simplified to array

    - -

    Value

    - -

    collect return a data.frame/data.table

    -

    collect_list returns a list

    -

    collect return a data.frame/data.table

    - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-# use collect to bring the data into RAM as a data.table/data.frame
-collect(cars.df)
-
    #>     speed dist
-#>  1:     4    2
-#>  2:     4   10
-#>  3:     7    4
-#>  4:     7   22
-#>  5:     8   16
-#>  6:     9   10
-#>  7:    10   18
-#>  8:    10   26
-#>  9:    10   34
-#> 10:    11   17
-#> 11:    11   28
-#> 12:    12   14
-#> 13:    12   20
-#> 14:    12   24
-#> 15:    12   28
-#> 16:    13   26
-#> 17:    13   34
-#> 18:    13   34
-#> 19:    13   46
-#> 20:    14   26
-#> 21:    14   36
-#> 22:    14   60
-#> 23:    14   80
-#> 24:    15   20
-#> 25:    15   26
-#> 26:    15   54
-#> 27:    16   32
-#> 28:    16   40
-#> 29:    17   32
-#> 30:    17   40
-#> 31:    17   50
-#> 32:    18   42
-#> 33:    18   56
-#> 34:    18   76
-#> 35:    18   84
-#> 36:    19   36
-#> 37:    19   46
-#> 38:    19   68
-#> 39:    20   32
-#> 40:    20   48
-#> 41:    20   52
-#> 42:    20   56
-#> 43:    20   64
-#> 44:    22   66
-#> 45:    23   54
-#> 46:    24   70
-#> 47:    24   92
-#> 48:    24   93
-#> 49:    24  120
-#> 50:    25   85
-#>     speed dist
    
-# clean up
-delete(cars.df)
-cars.df = as.disk.frame(cars)
+hence parallel = FALSE is a better choice
    
+
    simplify
    
+
    Should the result be simplified to array
    
+
    
+    
    
+    
    Value
    
+    
    collect return a data.frame/data.table
+collect_list returns a list
+collect return a data.frame/data.table
    
+    
    
 
-# returns the result as a list
-collect_list(cmap(cars.df, ~1))
-
    #> [[1]] -#> [1] 1 -#> -#> [[2]] -#> [1] 1 -#> -#> [[3]] -#> [1] 1 -#> -#> [[4]] -#> [1] 1 -#> -#> [[5]] -#> [1] 1 -#> -#> [[6]] -#> [1] 1 -#>
    -# clean up -delete(cars.df) -cars.df = as.disk.frame(cars) -# use collect to bring the data into RAM as a data.table/data.frame -collect(cars.df) -
    #> speed dist -#> 1: 4 2 -#> 2: 4 10 -#> 3: 7 4 -#> 4: 7 22 -#> 5: 8 16 -#> 6: 9 10 -#> 7: 10 18 -#> 8: 10 26 -#> 9: 10 34 -#> 10: 11 17 -#> 11: 11 28 -#> 12: 12 14 -#> 13: 12 20 -#> 14: 12 24 -#> 15: 12 28 -#> 16: 13 26 -#> 17: 13 34 -#> 18: 13 34 -#> 19: 13 46 -#> 20: 14 26 -#> 21: 14 36 -#> 22: 14 60 -#> 23: 14 80 -#> 24: 15 20 -#> 25: 15 26 -#> 26: 15 54 -#> 27: 16 32 -#> 28: 16 40 -#> 29: 17 32 -#> 30: 17 40 -#> 31: 17 50 -#> 32: 18 42 -#> 33: 18 56 -#> 34: 18 76 -#> 35: 18 84 -#> 36: 19 36 -#> 37: 19 46 -#> 38: 19 68 -#> 39: 20 32 -#> 40: 20 48 -#> 41: 20 52 -#> 42: 20 56 -#> 43: 20 64 -#> 44: 22 66 -#> 45: 23 54 -#> 46: 24 70 -#> 47: 24 92 -#> 48: 24 93 -#> 49: 24 120 -#> 50: 25 85 -#> speed dist
    -# clean up -delete(cars.df) -
    +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+# use collect to bring the data into RAM as a data.table/data.frame
+collect(cars.df)
+#>     speed dist
+#>  1:     4    2
+#>  2:     4   10
+#>  3:     7    4
+#>  4:     7   22
+#>  5:     8   16
+#>  6:     9   10
+#>  7:    10   18
+#>  8:    10   26
+#>  9:    10   34
+#> 10:    11   17
+#> 11:    11   28
+#> 12:    12   14
+#> 13:    12   20
+#> 14:    12   24
+#> 15:    12   28
+#> 16:    13   26
+#> 17:    13   34
+#> 18:    13   34
+#> 19:    13   46
+#> 20:    14   26
+#> 21:    14   36
+#> 22:    14   60
+#> 23:    14   80
+#> 24:    15   20
+#> 25:    15   26
+#> 26:    15   54
+#> 27:    16   32
+#> 28:    16   40
+#> 29:    17   32
+#> 30:    17   40
+#> 31:    17   50
+#> 32:    18   42
+#> 33:    18   56
+#> 34:    18   76
+#> 35:    18   84
+#> 36:    19   36
+#> 37:    19   46
+#> 38:    19   68
+#> 39:    20   32
+#> 40:    20   48
+#> 41:    20   52
+#> 42:    20   56
+#> 43:    20   64
+#> 44:    22   66
+#> 45:    23   54
+#> 46:    24   70
+#> 47:    24   92
+#> 48:    24   93
+#> 49:    24  120
+#> 50:    25   85
+#>     speed dist
+
+# clean up
+delete(cars.df)
+cars.df = as.disk.frame(cars)
+
+# returns the result as a list
+collect_list(cmap(cars.df, ~1))
+#> [[1]]
+#> [1] 1
+#> 
+#> [[2]]
+#> [1] 1
+#> 
+#> [[3]]
+#> [1] 1
+#> 
+#> [[4]]
+#> [1] 1
+#> 
+#> [[5]]
+#> [1] 1
+#> 
+#> [[6]]
+#> [1] 1
+#> 
+
+# clean up
+delete(cars.df)
+cars.df = as.disk.frame(cars)
+# use collect to bring the data into RAM as a data.table/data.frame
+collect(cars.df)
+#>     speed dist
+#>  1:     4    2
+#>  2:     4   10
+#>  3:     7    4
+#>  4:     7   22
+#>  5:     8   16
+#>  6:     9   10
+#>  7:    10   18
+#>  8:    10   26
+#>  9:    10   34
+#> 10:    11   17
+#> 11:    11   28
+#> 12:    12   14
+#> 13:    12   20
+#> 14:    12   24
+#> 15:    12   28
+#> 16:    13   26
+#> 17:    13   34
+#> 18:    13   34
+#> 19:    13   46
+#> 20:    14   26
+#> 21:    14   36
+#> 22:    14   60
+#> 23:    14   80
+#> 24:    15   20
+#> 25:    15   26
+#> 26:    15   54
+#> 27:    16   32
+#> 28:    16   40
+#> 29:    17   32
+#> 30:    17   40
+#> 31:    17   50
+#> 32:    18   42
+#> 33:    18   56
+#> 34:    18   76
+#> 35:    18   84
+#> 36:    19   36
+#> 37:    19   46
+#> 38:    19   68
+#> 39:    20   32
+#> 40:    20   48
+#> 41:    20   52
+#> 42:    20   56
+#> 43:    20   64
+#> 44:    22   66
+#> 45:    23   54
+#> 46:    24   70
+#> 47:    24   92
+#> 48:    24   93
+#> 49:    24  120
+#> 50:    25   85
+#>     speed dist
+
+# clean up
+delete(cars.df)
+
    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/colnames.html b/docs/reference/colnames.html index 432d0801..3743a499 100644 --- a/docs/reference/colnames.html +++ b/docs/reference/colnames.html @@ -1,70 +1,15 @@ - - - - - - - -Return the column names of the disk.frame — colnames • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Return the column names of the disk.frame — colnames • disk.frame - - - - - - - - - - - - - + + -
    -
    - -
    - -
    +

    Return the column names of the disk.frame

    - Source: R/names.r + Source: R/names.r
    @@ -171,56 +101,48 @@

    Return the column names of the disk.frame

    might take some time.

    - 
    colnames(x, ...)
+    
    
+    
    colnames(x, ...)
 
-# S3 method for disk.frame
-names(x, ...)
+# S3 method for disk.frame
+names(x, ...)
 
-# S3 method for disk.frame
-colnames(x, ...)
+# S3 method for disk.frame
+colnames(x, ...)
 
-# S3 method for default
-colnames(x, ...)
    
-
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
    x
    a disk.frame
    ...
    not used
    
+# S3 method for default
+colnames(x, ...)
    +
    +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    ...
    +

    not used

    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/compute.disk.frame.html b/docs/reference/compute.disk.frame.html index 9c62247e..8feaf6eb 100644 --- a/docs/reference/compute.disk.frame.html +++ b/docs/reference/compute.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Compute without writing — compute.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Compute without writing — compute.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Compute without writing

    - Source: R/compute.r + Source: R/compute.r
    @@ -165,76 +95,64 @@

    Compute without writing

    Perform the computation; same as calling cmap without .f and lazy = FALSE

    - 
    # S3 method for disk.frame
-compute(
-  x,
-  name,
-  outdir = tempfile("tmp_df_", fileext = ".df"),
-  overwrite = TRUE,
-  ...
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - -
    x

    a disk.frame

    name

    Not used. Kept for compatibility with dplyr

    outdir

    the output directory

    overwrite

    whether to overwrite or not

    ...

    Not used. Kept for dplyr compatibility

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-cars.df2 = cars.df %>% cmap(~.x)
-# the computation is performed and the data is now stored elsewhere
-cars.df3 = compute(cars.df2)
-
-# clean up
-delete(cars.df)
-delete(cars.df3)
-
    +
    + 
    # S3 method for disk.frame
+compute(
+  x,
+  name,
+  outdir = tempfile("tmp_df_", fileext = ".df"),
+  overwrite = TRUE,
+  ...
+)
    +
    + +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    name
    +

    Not used. Kept for compatibility with dplyr

    +
    outdir
    +

    the output directory

    +
    overwrite
    +

    whether to overwrite or not

    +
    ...
    +

    Not used. Kept for dplyr compatibility

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+cars.df2 = cars.df %>% cmap(~.x)
+# the computation is performed and the data is now stored elsewhere
+cars.df3 = compute(cars.df2)
+
+# clean up
+delete(cars.df)
+delete(cars.df3)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/create_chunk_mapper.html b/docs/reference/create_chunk_mapper.html index 041c6bb6..7c28d311 100644 --- a/docs/reference/create_chunk_mapper.html +++ b/docs/reference/create_chunk_mapper.html @@ -1,67 +1,12 @@ - - - - - - - -Create function that applies to each chunk if disk.frame — create_chunk_mapper • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Create function that applies to each chunk if disk.frame — create_chunk_mapper • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Create function that applies to each chunk if disk.frame

    - Source: R/chunk_mapper.r + Source: R/chunk_mapper.r
    @@ -165,108 +95,102 @@

    Create function that applies to each chunk if disk.frame

    A function to make it easier to create functions like filter

    - 
    create_chunk_mapper(chunk_fn, warning_msg = NULL, as.data.frame = TRUE)
    - -

    Arguments

    - - - - - - - - - - - - - - -
    chunk_fn

    The dplyr function to create a mapper for

    warning_msg

    The warning message to display when invoking the mapper

    as.data.frame

    force the input chunk of a data.frame; needed for dtplyr

    - - -

    Examples

    - 
    
-filter = create_chunk_mapper(dplyr::filter)
-
-#' example: creating a function that keeps only the first and last n row
-first_and_last <- function(chunk, n, ...) {
-  nr = nrow(chunk)
-  print(nr-n+1:nr)
-  chunk[c(1:n, (nr-n+1):nr), ]
-}
-
-#' create the function for use with disk.frame
-first_and_last_df = create_chunk_mapper(first_and_last)
-
-mtcars.df = as.disk.frame(mtcars)
-
-#' the operation is lazy
-lazy_mtcars.df = mtcars.df %>%
-  first_and_last_df(2)
-
-#' bring into R
-collect(lazy_mtcars.df)
-
    #> [1]  5  6  7  8  9 10
-#> [1]  5  6  7  8  9 10
-#> [1]  5  6  7  8  9 10
-#> [1]  5  6  7  8  9 10
-#> [1]  5  6  7  8  9 10
-#> [1] 1 2
    #>      mpg cyl  disp  hp drat    wt  qsec vs am gear carb
-#>  1: 21.0   6 160.0 110 3.90 2.620 16.46  0  1    4    4
-#>  2: 21.0   6 160.0 110 3.90 2.875 17.02  0  1    4    4
-#>  3: 18.7   8 360.0 175 3.15 3.440 17.02  0  0    3    2
-#>  4: 18.1   6 225.0 105 2.76 3.460 20.22  1  0    3    1
-#>  5: 14.3   8 360.0 245 3.21 3.570 15.84  0  0    3    4
-#>  6: 24.4   4 146.7  62 3.69 3.190 20.00  1  0    4    2
-#>  7: 17.8   6 167.6 123 3.92 3.440 18.90  1  0    4    4
-#>  8: 16.4   8 275.8 180 3.07 4.070 17.40  0  0    3    3
-#>  9: 17.3   8 275.8 180 3.07 3.730 17.60  0  0    3    3
-#> 10: 15.2   8 275.8 180 3.07 3.780 18.00  0  0    3    3
-#> 11: 14.7   8 440.0 230 3.23 5.345 17.42  0  0    3    4
-#> 12: 32.4   4  78.7  66 4.08 2.200 19.47  1  1    4    1
-#> 13: 30.4   4  75.7  52 4.93 1.615 18.52  1  1    4    2
-#> 14: 33.9   4  71.1  65 4.22 1.835 19.90  1  1    4    1
-#> 15: 15.2   8 304.0 150 3.15 3.435 17.30  0  0    3    2
-#> 16: 13.3   8 350.0 245 3.73 3.840 15.41  0  0    3    4
-#> 17: 19.2   8 400.0 175 3.08 3.845 17.05  0  0    3    2
-#> 18: 27.3   4  79.0  66 4.08 1.935 18.90  1  1    4    1
-#> 19: 15.8   8 351.0 264 4.22 3.170 14.50  0  1    5    4
-#> 20: 19.7   6 145.0 175 3.62 2.770 15.50  0  1    5    6
-#> 21: 15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
-#> 22: 21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
-#> 23: 15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
-#> 24: 21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
-#>      mpg cyl  disp  hp drat    wt  qsec vs am gear carb
    
-#' clean up
-delete(mtcars.df)
+    
    
+    
    create_chunk_mapper(chunk_fn, warning_msg = NULL, as.data.frame = TRUE)
    
+    
    
 
-
    +
    +

    Arguments

    +
    chunk_fn
    +

    The dplyr function to create a mapper for

    +
    warning_msg
    +

    The warning message to display when invoking the mapper

    +
    as.data.frame
    +

    force the input chunk of a data.frame; needed for dtplyr

    +
    + +
    +

    Examples

    + 
    
+filter = create_chunk_mapper(dplyr::filter)
+
+#' example: creating a function that keeps only the first and last n row
+first_and_last <- function(chunk, n, ...) {
+  nr = nrow(chunk)
+  print(nr-n+1:nr)
+  chunk[c(1:n, (nr-n+1):nr), ]
+}
+
+#' create the function for use with disk.frame
+first_and_last_df = create_chunk_mapper(first_and_last)
+
+mtcars.df = as.disk.frame(mtcars)
+
+#' the operation is lazy
+lazy_mtcars.df = mtcars.df %>%
+  first_and_last_df(2)
+
+#' bring into R
+collect(lazy_mtcars.df)
+#> [1]  5  6  7  8  9 10
+#> [1]  5  6  7  8  9 10
+#> [1]  5  6  7  8  9 10
+#> [1]  5  6  7  8  9 10
+#> [1]  5  6  7  8  9 10
+#> [1] 1 2
+#>      mpg cyl  disp  hp drat    wt  qsec vs am gear carb
+#>  1: 21.0   6 160.0 110 3.90 2.620 16.46  0  1    4    4
+#>  2: 21.0   6 160.0 110 3.90 2.875 17.02  0  1    4    4
+#>  3: 18.7   8 360.0 175 3.15 3.440 17.02  0  0    3    2
+#>  4: 18.1   6 225.0 105 2.76 3.460 20.22  1  0    3    1
+#>  5: 14.3   8 360.0 245 3.21 3.570 15.84  0  0    3    4
+#>  6: 24.4   4 146.7  62 3.69 3.190 20.00  1  0    4    2
+#>  7: 17.8   6 167.6 123 3.92 3.440 18.90  1  0    4    4
+#>  8: 16.4   8 275.8 180 3.07 4.070 17.40  0  0    3    3
+#>  9: 17.3   8 275.8 180 3.07 3.730 17.60  0  0    3    3
+#> 10: 15.2   8 275.8 180 3.07 3.780 18.00  0  0    3    3
+#> 11: 14.7   8 440.0 230 3.23 5.345 17.42  0  0    3    4
+#> 12: 32.4   4  78.7  66 4.08 2.200 19.47  1  1    4    1
+#> 13: 30.4   4  75.7  52 4.93 1.615 18.52  1  1    4    2
+#> 14: 33.9   4  71.1  65 4.22 1.835 19.90  1  1    4    1
+#> 15: 15.2   8 304.0 150 3.15 3.435 17.30  0  0    3    2
+#> 16: 13.3   8 350.0 245 3.73 3.840 15.41  0  0    3    4
+#> 17: 19.2   8 400.0 175 3.08 3.845 17.05  0  0    3    2
+#> 18: 27.3   4  79.0  66 4.08 1.935 18.90  1  1    4    1
+#> 19: 15.8   8 351.0 264 4.22 3.170 14.50  0  1    5    4
+#> 20: 19.7   6 145.0 175 3.62 2.770 15.50  0  1    5    6
+#> 21: 15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
+#> 22: 21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
+#> 23: 15.0   8 301.0 335 3.54 3.570 14.60  0  1    5    8
+#> 24: 21.4   4 121.0 109 4.11 2.780 18.60  1  1    4    2
+#>      mpg cyl  disp  hp drat    wt  qsec vs am gear carb
+
+#' clean up
+delete(mtcars.df)
+
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/csv_to_disk.frame.html b/docs/reference/csv_to_disk.frame.html index 687aa9d8..17b0d387 100644 --- a/docs/reference/csv_to_disk.frame.html +++ b/docs/reference/csv_to_disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Convert CSV file(s) to disk.frame format — csv_to_disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Convert CSV file(s) to disk.frame format — csv_to_disk.frame • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Convert CSV file(s) to disk.frame format

    - Source: R/csv2disk.frame.r + Source: R/csv2disk.frame.r
    @@ -165,142 +95,115 @@

    Convert CSV file(s) to disk.frame format

    Convert CSV file(s) to disk.frame format

    - 
    csv_to_disk.frame(
-  infile,
-  outdir = tempfile(fileext = ".df"),
-  inmapfn = base::I,
-  nchunks = recommend_nchunks(sum(file.size(infile))),
-  in_chunk_size = NULL,
-  shardby = NULL,
-  compress = 50,
-  overwrite = TRUE,
-  header = TRUE,
-  .progress = TRUE,
-  backend = c("data.table", "readr", "LaF"),
-  chunk_reader = c("bigreadr", "data.table", "readr", "readLines"),
-  ...
-)
    +
    + 
    csv_to_disk.frame(
+  infile,
+  outdir = tempfile(fileext = ".df"),
+  inmapfn = base::I,
+  nchunks = recommend_nchunks(sum(file.size(infile))),
+  in_chunk_size = NULL,
+  shardby = NULL,
+  compress = 50,
+  overwrite = TRUE,
+  header = TRUE,
+  .progress = TRUE,
+  backend = c("data.table", "readr", "LaF"),
+  chunk_reader = c("bigreadr", "data.table", "readr", "readLines"),
+  ...
+)
    +
    -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    infile

    The input CSV file or files

    outdir

    The directory to output the disk.frame to

    inmapfn

    A function to be applied to the chunk read in from CSV before +

    +

    Arguments

    +
    infile
    +

    The input CSV file or files

    +
    outdir
    +

    The directory to output the disk.frame to

    +
    inmapfn
    +

    A function to be applied to the chunk read in from CSV before the chunk is being written out. Commonly used to perform simple -transformations. Defaults to the identity function (ie. no transformation)

    nchunks

    Number of chunks to output

    in_chunk_size

    When reading in the file, how many lines to read in at +transformations. Defaults to the identity function (ie. no transformation)

    +
    nchunks
    +

    Number of chunks to output

    +
    in_chunk_size
    +

    When reading in the file, how many lines to read in at once. This is different to nchunks which controls how many chunks are -output

    shardby

    The column(s) to shard the data by. For example suppose +output

    +
    shardby
    +

    The column(s) to shard the data by. For example suppose `shardby = c("col1","col2")` then every row where the values `col1` and `col2` are the same will end up in the same chunk; this will allow merging -by `col1` and `col2` to be more efficient

    compress

    For fst backends it's a number between 0 and 100 where 100 is -the highest compression ratio.

    overwrite

    Whether to overwrite the existing directory

    header

    Whether the files have header. Defaults to TRUE

    .progress

    A logical, for whether or not to print a progress bar for -multiprocess, multisession, and multicore plans. From furrr

    backend

    The CSV reader backend to choose: "data.table" or "readr". +by `col1` and `col2` to be more efficient

    +
    compress
    +

    For fst backends it's a number between 0 and 100 where 100 is +the highest compression ratio.

    +
    overwrite
    +

    Whether to overwrite the existing directory

    +
    header
    +

    Whether the files have header. Defaults to TRUE

    +
    .progress
    +

    A logical, for whether or not to print a progress bar for +multiprocess, multisession, and multicore plans. From furrr

    +
    backend
    +

    The CSV reader backend to choose: "data.table" or "readr". disk.frame does not have its own CSV reader. It uses either data.table::fread or readr::read_delimited. It is worth noting that data.table::fread does not detect dates and all dates are imported as strings, and you are encouraged to use fasttime to convert the strings to date. You can use the `inmapfn` to do that. However, if you want automatic date detection, then backend="readr" may suit your needs. However, readr -is often slower than data.table, hence data.table is chosen as the default.

    chunk_reader

    Even if you choose a backend there can still be multiple +is often slower than data.table, hence data.table is chosen as the default.

    +
    chunk_reader
    +

    Even if you choose a backend there can still be multiple strategies on how to approach the CSV reads. For example, data.table::fread tries to mmap the whole file which can cause the whole read process to fail. In that case we can change the chunk_reader to "readLines" which uses the readLines function to read chunk by chunk and still use data.table::fread to process the chunks. There are currently no strategies for readr backend, -except the default one.

    ...

    passed to data.table::fread, disk.frame::as.disk.frame, -disk.frame::shard

    - -

    See also

    - -

    Other ingesting data: -zip_to_disk.frame()

    - -

    Examples

    - 
    tmpfile = tempfile()
-write.csv(cars, tmpfile)
-tmpdf = tempfile(fileext = ".df")
-df = csv_to_disk.frame(tmpfile, outdir = tmpdf, overwrite = TRUE)
+except the default one.
    
+
    ...
    
+
    passed to data.table::fread, disk.frame::as.disk.frame,
+disk.frame::shard
    
+
    
+    
    
+    
    See also
    
+    
    Other ingesting data: 
+zip_to_disk.frame()
    
+    
    
 
-# clean up
-fs::file_delete(tmpfile)
-delete(df)
-
    +
    +

    Examples

    + 
    tmpfile = tempfile()
+write.csv(cars, tmpfile)
+tmpdf = tempfile(fileext = ".df")
+df = csv_to_disk.frame(tmpfile, outdir = tmpdf, overwrite = TRUE)
+
+# clean up
+fs::file_delete(tmpfile)
+delete(df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/delete.html b/docs/reference/delete.html index d890c7ba..2a238505 100644 --- a/docs/reference/delete.html +++ b/docs/reference/delete.html @@ -1,67 +1,12 @@ - - - - - - - -Delete a disk.frame — delete • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Delete a disk.frame — delete • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Delete a disk.frame

    - Source: R/delete.r + Source: R/delete.r
    @@ -165,47 +95,43 @@

    Delete a disk.frame

    Delete a disk.frame

    - 
    delete(df)
    - -

    Arguments

    - - - - - - -
    df

    a disk.frame

    - +
    + 
    delete(df)
    +
    -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-delete(cars.df)
-
    +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/df_ram_size.html b/docs/reference/df_ram_size.html index c3ac569a..57bfaa8d 100644 --- a/docs/reference/df_ram_size.html +++ b/docs/reference/df_ram_size.html @@ -1,67 +1,12 @@ - - - - - - - -Get the size of RAM in gigabytes — df_ram_size • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Get the size of RAM in gigabytes — df_ram_size • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Get the size of RAM in gigabytes

    - Source: R/recommend_nchunks.r + Source: R/recommend_nchunks.r
    @@ -165,42 +95,43 @@

    Get the size of RAM in gigabytes

    Get the size of RAM in gigabytes

    - 
    df_ram_size()
    - - -

    Value

    +
    + 
    df_ram_size()
    +
    +
    +

    Value

    integer of RAM in gigabyte (GB)

    +
    -

    Examples

    - 
    # returns the RAM size in gigabyte (GB)
-df_ram_size() 
-
    #> [1] 64
    +
    +

    Examples

    + 
    # returns the RAM size in gigabyte (GB)
+df_ram_size() 
+#> [1] 64
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/dfglm.html b/docs/reference/dfglm.html index ee6fc36b..1197ac98 100644 --- a/docs/reference/dfglm.html +++ b/docs/reference/dfglm.html @@ -1,68 +1,13 @@ - - - - - - - -Fit generalized linear models (glm) with disk.frame — dfglm • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Fit generalized linear models (glm) with disk.frame — dfglm • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Fit generalized linear models (glm) with disk.frame

    - Source: R/glm.r + Source: R/glm.r
    @@ -167,37 +97,30 @@

    Fit generalized linear models (glm) with disk.frame

    those return by those functions. This is a convenience wrapper

    - 
    dfglm(formula, data, ..., glm_backend = c("biglm", "speedglm", "biglmm"))
    +
    + 
    dfglm(formula, data, ..., glm_backend = c("biglm", "speedglm", "biglmm"))
    +
    -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    formula

    A model formula

    data

    See Details below. Method dispatch is on this argument

    ...

    Additional arguments

    glm_backend

    Which package to use for fitting GLMs. The default is +

    +

    Arguments

    +
    formula
    +

    A model formula

    +
    data
    +

    See Details below. Method dispatch is on this argument

    +
    ...
    +

    Additional arguments

    +
    glm_backend
    +

    Which package to use for fitting GLMs. The default is "biglm", which has known issues with factor level if different levels are present in different chunks. The "speedglm" option is more robust, but does not -implement `predict` which makes prediction and implementation impossible.

    - -

    Value

    - +implement `predict` which makes prediction and implementation impossible.

    +
    +
    +

    Value

    An object of class bigglm

    -

    Details

    - +
    +
    +

    Details

    The data argument may be a function, a data frame, or a SQLiteConnection or RODBC connection object.

    When it is a function the function must take a single argument @@ -217,63 +140,66 @@

    Details variables needed for the model, not the whole table. The code in the SQLiteConnection method should work for other DBI connections, but I do not have any of these to check it with.

    -

    References

    - +
    +
    +

    References

    Algorithm AS274 Applied Statistics (1992) Vol.41, No. 2

    -

    See also

    - -

    Other Machine Learning (ML): -make_glm_streaming_fn()

    +
    +
    +

    See also

    +

    Other Machine Learning (ML): +make_glm_streaming_fn()

    +
    -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-m = dfglm(dist ~ speed, data = cars.df)
-
    #> Loading required namespace: biglm
    
-# can use normal R functions
-# Only works in version > R 3.6
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
-if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
-  summary(m)
-  predict(m, get_chunk(cars.df, 1))
-  predict(m, collect(cars.df))
-  # can use broom to tidy up the returned info
-  broom::tidy(m)
-}
-
    #> # A tibble: 2 x 4
-#>   term        estimate std.error  p.value
-#>   <chr>          <dbl>     <dbl>    <dbl>
-#> 1 (Intercept)   -17.6      6.76  9.29e- 3
-#> 2 speed           3.93     0.416 2.96e-21
    
-# clean up
-delete(cars.df)
-
    +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+m = dfglm(dist ~ speed, data = cars.df)
+#> Loading required namespace: biglm
+
+# can use normal R functions
+# Only works in version > R 3.6
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
+  summary(m)
+  predict(m, get_chunk(cars.df, 1))
+  predict(m, collect(cars.df))
+  # can use broom to tidy up the returned info
+  broom::tidy(m)
+}
+#> # A tibble: 2 x 4
+#>   term        estimate std.error  p.value
+#>   <chr>          <dbl>     <dbl>    <dbl>
+#> 1 (Intercept)   -17.6      6.76  9.29e- 3
+#> 2 speed           3.93     0.416 2.96e-21
+
+# clean up
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/disk.frame.html b/docs/reference/disk.frame.html index 574c9440..0d2320c2 100644 --- a/docs/reference/disk.frame.html +++ b/docs/reference/disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Create a disk.frame from a folder — disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Create a disk.frame from a folder — disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Create a disk.frame from a folder

    - Source: R/disk.frame.r + Source: R/disk.frame.r
    @@ -165,67 +95,64 @@

    Create a disk.frame from a folder

    Create a disk.frame from a folder

    - 
    disk.frame(path, backend = "fst")
    - -

    Arguments

    - - - - - - - - - - -
    path

    The path to store the output file or to a directory

    backend

    The only available backend is fst at the moment

    - - -

    Examples

    - 
    path = file.path(tempdir(),"cars")
-as.disk.frame(cars, outdir=path, overwrite = TRUE, nchunks = 2)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP/cars"
-#> nchunks: 2
-#> nrow (at source): 50
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    df = disk.frame(path)
-head(df)
-
    #>    speed dist
-#> 1:     4    2
-#> 2:     4   10
-#> 3:     7    4
-#> 4:     7   22
-#> 5:     8   16
-#> 6:     9   10
    nchunks(df)
-
    #> [1] 2
    # clean up
-delete(df)
-
    +
    + 
    disk.frame(path, backend = "fst")
    +
    + +
    +

    Arguments

    +
    path
    +

    The path to store the output file or to a directory

    +
    backend
    +

    The only available backend is fst at the moment

    +
    + +
    +

    Examples

    + 
    path = file.path(tempdir(),"cars")
+as.disk.frame(cars, outdir=path, overwrite = TRUE, nchunks = 2)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5/cars"
+#> nchunks: 2
+#> nrow (at source): 50
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+df = disk.frame(path)
+head(df)
+#>    speed dist
+#> 1:     4    2
+#> 2:     4   10
+#> 3:     7    4
+#> 4:     7   22
+#> 5:     8   16
+#> 6:     9   10
+nchunks(df)
+#> [1] 2
+# clean up
+delete(df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/dplyr_verbs.html b/docs/reference/dplyr_verbs.html index 9f86f10e..4cbb0612 100644 --- a/docs/reference/dplyr_verbs.html +++ b/docs/reference/dplyr_verbs.html @@ -1,68 +1,13 @@ - - - - - - - -The dplyr verbs implemented for disk.frame — select.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -The dplyr verbs implemented for disk.frame — select.disk.frame • disk.frame + + - - - - -
    -
    - -
    - -
    +

    The dplyr verbs implemented for disk.frame

    - Source: R/dplyr_verbs.r + Source: R/dplyr_verbs.r
    @@ -167,94 +97,88 @@

    The dplyr verbs implemented for disk.frame

    `chunk_arrange` performs the actions within each chunk

    - 
    # S3 method for disk.frame
-select(.data, ...)
-
-# S3 method for disk.frame
-rename(.data, ...)
-
-# S3 method for disk.frame
-filter(.data, ...)
-
-# S3 method for disk.frame
-mutate(.data, ...)
-
-# S3 method for disk.frame
-transmute(.data, ...)
+    
    
+    
    # S3 method for disk.frame
+select(.data, ...)
 
-# S3 method for disk.frame
-arrange(.data, ...)
+# S3 method for disk.frame
+rename(.data, ...)
 
-chunk_arrange(.data, ...)
+# S3 method for disk.frame
+filter(.data, ...)
 
-add_tally.disk.frame(.data, ...)
+# S3 method for disk.frame
+mutate(.data, ...)
 
-# S3 method for disk.frame
-do(.data, ...)
+# S3 method for disk.frame
+transmute(.data, ...)
 
-# S3 method for disk.frame
-distinct(...)
+# S3 method for disk.frame
+arrange(.data, ...)
 
-chunk_distinct(.data, ...)
+chunk_arrange(.data, ...)
 
-# S3 method for disk.frame
-glimpse(.data, ...)
    
+add_tally.disk.frame(.data, ...)
 
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
    .data
    a disk.frame
    ...
    Same as the dplyr functions
    
+# S3 method for disk.frame
+do(.data, ...)
 
+# S3 method for disk.frame
+distinct(...)
 
-    
    Examples
    
-    
    library(dplyr)
-cars.df = as.disk.frame(cars)
-mult = 2
+chunk_distinct(.data, ...)
 
-# use all any of the supported dplyr
-cars2 = cars.df %>% 
-  select(speed) %>% 
-  mutate(speed2 = speed * mult) %>% 
-  filter(speed < 50) %>% 
-  rename(speed1 = speed) %>% 
-  collect
+# S3 method for disk.frame
+glimpse(.data, ...)
    
+    
    
 
-# clean up cars.df
-delete(cars.df)
-
    +
    +

    Arguments

    +
    .data
    +

    a disk.frame

    +
    ...
    +

    Same as the dplyr functions

    +
    + +
    +

    Examples

    + 
    library(dplyr)
+cars.df = as.disk.frame(cars)
+mult = 2
+
+# use all any of the supported dplyr
+cars2 = cars.df %>% 
+  select(speed) %>% 
+  mutate(speed2 = speed * mult) %>% 
+  filter(speed < 50) %>% 
+  rename(speed1 = speed) %>% 
+  collect
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/evalparseglue.html b/docs/reference/evalparseglue.html index ef02aa33..4172714e 100644 --- a/docs/reference/evalparseglue.html +++ b/docs/reference/evalparseglue.html @@ -1,67 +1,12 @@ - - - - - - - -Helper function to evalparse some `glue::glue` string — evalparseglue • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Helper function to evalparse some `glue::glue` string — evalparseglue • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Helper function to evalparse some `glue::glue` string

    - Source: R/util.r + Source: R/util.r
    @@ -165,47 +95,39 @@

    Helper function to evalparse some `glue::glue` string

    Helper function to evalparse some `glue::glue` string

    - 
    evalparseglue(code, env = parent.frame())
    - -

    Arguments

    - - - - - - - - - - -
    code

    the code in character(string) format to evaluate

    env

    the environment in which to evaluate the code

    +
    + 
    evalparseglue(code, env = parent.frame())
    +
    +
    +

    Arguments

    +
    code
    +

    the code in character(string) format to evaluate

    +
    env
    +

    the environment in which to evaluate the code

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/foverlaps.disk.frame.html b/docs/reference/foverlaps.disk.frame.html index df3534db..19607792 100644 --- a/docs/reference/foverlaps.disk.frame.html +++ b/docs/reference/foverlaps.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Apply data.table's foverlaps to the disk.frame — foverlaps.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Apply data.table's foverlaps to the disk.frame — foverlaps.disk.frame • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Apply data.table's foverlaps to the disk.frame

    - Source: R/foverlaps.disk.frame.r + Source: R/foverlaps.disk.frame.r
    @@ -165,101 +95,81 @@

    Apply data.table's foverlaps to the disk.frame

    EXPERIMENTAL

    - 
    foverlaps.disk.frame(
-  df1,
-  df2,
-  by.x = if (identical(shardkey(df1)$shardkey, "")) shardkey(df1)$shardkey else
-    shardkey(df2)$shardkey,
-  by.y = shardkey(df2)$shardkey,
-  ...,
-  outdir = tempfile("df_foverlaps_tmp", fileext = ".df"),
-  merge_by_chunk_id = FALSE,
-  compress = 50,
-  overwrite = TRUE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df1

    A disk.frame

    df2

    A disk.frame or a data.frame

    by.x

    character/string vector. by.x used in foverlaps

    by.y

    character/string vector. by.x used in foverlaps

    ...

    passed to data.table::foverlaps and disk.frame::cmap.disk.frame

    outdir

    The output directory of the disk.frame

    merge_by_chunk_id

    If TRUE then the merges will happen for chunks in df1 and df2 with the same chunk id which speed up processing. Otherwise every chunk of df1 is merged with every chunk of df2. Ignored with df2 is not a disk.frame

    compress

    The compression ratio for fst

    overwrite

    overwrite existing directory

    - - -

    Examples

    - 
    library(data.table)
+    
    
+    
    foverlaps.disk.frame(
+  df1,
+  df2,
+  by.x = if (identical(shardkey(df1)$shardkey, "")) shardkey(df1)$shardkey else
+    shardkey(df2)$shardkey,
+  by.y = shardkey(df2)$shardkey,
+  ...,
+  outdir = tempfile("df_foverlaps_tmp", fileext = ".df"),
+  merge_by_chunk_id = FALSE,
+  compress = 50,
+  overwrite = TRUE
+)
    
+    
    
 
-## simple example:
-x = as.disk.frame(data.table(start=c(5,31,22,16), end=c(8,50,25,18), val2 = 7:10))
-y = as.disk.frame(data.table(start=c(10, 20, 30), end=c(15, 35, 45), val1 = 1:3))
-byxy = c("start", "end")
-xy.df = foverlaps.disk.frame(
-   x, y, by.x = byxy, by.y = byxy,
-  merge_by_chunk_id = TRUE, overwrite = TRUE)
-# clean up
-delete(x)
-delete(y)
-delete(xy.df)
-
    +
    +

    Arguments

    +
    df1
    +

    A disk.frame

    +
    df2
    +

    A disk.frame or a data.frame

    +
    by.x
    +

    character/string vector. by.x used in foverlaps

    +
    by.y
    +

    character/string vector. by.x used in foverlaps

    +
    ...
    +

    passed to data.table::foverlaps and disk.frame::cmap.disk.frame

    +
    outdir
    +

    The output directory of the disk.frame

    +
    merge_by_chunk_id
    +

    If TRUE then the merges will happen for chunks in df1 and df2 with the same chunk id which speed up processing. Otherwise every chunk of df1 is merged with every chunk of df2. Ignored with df2 is not a disk.frame

    +
    compress
    +

    The compression ratio for fst

    +
    overwrite
    +

    overwrite existing directory

    +
    + +
    +

    Examples

    + 
    library(data.table)
+
+## simple example:
+x = as.disk.frame(data.table(start=c(5,31,22,16), end=c(8,50,25,18), val2 = 7:10))
+y = as.disk.frame(data.table(start=c(10, 20, 30), end=c(15, 35, 45), val1 = 1:3))
+byxy = c("start", "end")
+xy.df = foverlaps.disk.frame(
+   x, y, by.x = byxy, by.y = byxy,
+  merge_by_chunk_id = TRUE, overwrite = TRUE)
+# clean up
+delete(x)
+delete(y)
+delete(xy.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/gen_datatable_synthetic.html b/docs/reference/gen_datatable_synthetic.html index 9a068fca..608041f6 100644 --- a/docs/reference/gen_datatable_synthetic.html +++ b/docs/reference/gen_datatable_synthetic.html @@ -1,67 +1,12 @@ - - - - - - - -Generate synthetic dataset for testing — gen_datatable_synthetic • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Generate synthetic dataset for testing — gen_datatable_synthetic • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Generate synthetic dataset for testing

    - Source: R/util.r + Source: R/util.r
    @@ -165,47 +95,39 @@

    Generate synthetic dataset for testing

    Generate synthetic dataset for testing

    - 
    gen_datatable_synthetic(N = 2e+08, K = 100)
    - -

    Arguments

    - - - - - - - - - - -
    N

    number of rows. Defaults to 200 million

    K

    controls the number of unique values for id. Some ids will have K distinct values while others have N/K distinct values

    +
    + 
    gen_datatable_synthetic(N = 2e+08, K = 100)
    +
    +
    +

    Arguments

    +
    N
    +

    number of rows. Defaults to 200 million

    +
    K
    +

    controls the number of unique values for id. Some ids will have K distinct values while others have N/K distinct values

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/get_chunk.html b/docs/reference/get_chunk.html index 517e982e..66df5c10 100644 --- a/docs/reference/get_chunk.html +++ b/docs/reference/get_chunk.html @@ -1,67 +1,12 @@ - - - - - - - -Obtain one chunk by chunk id — get_chunk • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Obtain one chunk by chunk id — get_chunk • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Obtain one chunk by chunk id

    - Source: R/get_chunk.r + Source: R/get_chunk.r
    @@ -165,178 +95,170 @@

    Obtain one chunk by chunk id

    Obtain one chunk by chunk id

    - 
    get_chunk(...)
-
-# S3 method for disk.frame
-get_chunk(df, n, keep = NULL, full.names = FALSE, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - -
    ...

    passed to fst::read_fst or whichever read function is used in the backend

    df

    a disk.frame

    n

    the chunk id. If numeric then matches by number, if character then returns the chunk with the same name as n

    keep

    the columns to keep

    full.names

    whether n is the full path to the chunks or just a relative path file name. Ignored if n is numeric

    +
    + 
    get_chunk(...)
 
+# S3 method for disk.frame
+get_chunk(df, n, keep = NULL, full.names = FALSE, ...)
    +
    -

    Examples

    - 
    cars.df = as.disk.frame(cars, nchunks = 2)
-get_chunk(cars.df, 1)
-
    #>     speed dist
-#>  1:     4    2
-#>  2:     4   10
-#>  3:     7    4
-#>  4:     7   22
-#>  5:     8   16
-#>  6:     9   10
-#>  7:    10   18
-#>  8:    10   26
-#>  9:    10   34
-#> 10:    11   17
-#> 11:    11   28
-#> 12:    12   14
-#> 13:    12   20
-#> 14:    12   24
-#> 15:    12   28
-#> 16:    13   26
-#> 17:    13   34
-#> 18:    13   34
-#> 19:    13   46
-#> 20:    14   26
-#> 21:    14   36
-#> 22:    14   60
-#> 23:    14   80
-#> 24:    15   20
-#> 25:    15   26
-#>     speed dist
    get_chunk(cars.df, 2)
-
    #>     speed dist
-#>  1:    15   54
-#>  2:    16   32
-#>  3:    16   40
-#>  4:    17   32
-#>  5:    17   40
-#>  6:    17   50
-#>  7:    18   42
-#>  8:    18   56
-#>  9:    18   76
-#> 10:    18   84
-#> 11:    19   36
-#> 12:    19   46
-#> 13:    19   68
-#> 14:    20   32
-#> 15:    20   48
-#> 16:    20   52
-#> 17:    20   56
-#> 18:    20   64
-#> 19:    22   66
-#> 20:    23   54
-#> 21:    24   70
-#> 22:    24   92
-#> 23:    24   93
-#> 24:    24  120
-#> 25:    25   85
-#>     speed dist
    get_chunk(cars.df, 1, keep = "speed")
-
    #>     speed
-#>  1:     4
-#>  2:     4
-#>  3:     7
-#>  4:     7
-#>  5:     8
-#>  6:     9
-#>  7:    10
-#>  8:    10
-#>  9:    10
-#> 10:    11
-#> 11:    11
-#> 12:    12
-#> 13:    12
-#> 14:    12
-#> 15:    12
-#> 16:    13
-#> 17:    13
-#> 18:    13
-#> 19:    13
-#> 20:    14
-#> 21:    14
-#> 22:    14
-#> 23:    14
-#> 24:    15
-#> 25:    15
-#>     speed
    
-# if full.names = TRUE then the full path to the chunk need to be provided
-get_chunk(cars.df, file.path(attr(cars.df, "path"), "1.fst"), full.names = TRUE)
-
    #>     speed dist
-#>  1:     4    2
-#>  2:     4   10
-#>  3:     7    4
-#>  4:     7   22
-#>  5:     8   16
-#>  6:     9   10
-#>  7:    10   18
-#>  8:    10   26
-#>  9:    10   34
-#> 10:    11   17
-#> 11:    11   28
-#> 12:    12   14
-#> 13:    12   20
-#> 14:    12   24
-#> 15:    12   28
-#> 16:    13   26
-#> 17:    13   34
-#> 18:    13   34
-#> 19:    13   46
-#> 20:    14   26
-#> 21:    14   36
-#> 22:    14   60
-#> 23:    14   80
-#> 24:    15   20
-#> 25:    15   26
-#>     speed dist
    
-# clean up cars.df
-delete(cars.df)
-
    +
    +

    Arguments

    +
    ...
    +

    passed to fst::read_fst or whichever read function is used in the backend

    +
    df
    +

    a disk.frame

    +
    n
    +

    the chunk id. If numeric then matches by number, if character then returns the chunk with the same name as n

    +
    keep
    +

    the columns to keep

    +
    full.names
    +

    whether n is the full path to the chunks or just a relative path file name. Ignored if n is numeric

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars, nchunks = 2)
+get_chunk(cars.df, 1)
+#>     speed dist
+#>  1:     4    2
+#>  2:     4   10
+#>  3:     7    4
+#>  4:     7   22
+#>  5:     8   16
+#>  6:     9   10
+#>  7:    10   18
+#>  8:    10   26
+#>  9:    10   34
+#> 10:    11   17
+#> 11:    11   28
+#> 12:    12   14
+#> 13:    12   20
+#> 14:    12   24
+#> 15:    12   28
+#> 16:    13   26
+#> 17:    13   34
+#> 18:    13   34
+#> 19:    13   46
+#> 20:    14   26
+#> 21:    14   36
+#> 22:    14   60
+#> 23:    14   80
+#> 24:    15   20
+#> 25:    15   26
+#>     speed dist
+get_chunk(cars.df, 2)
+#>     speed dist
+#>  1:    15   54
+#>  2:    16   32
+#>  3:    16   40
+#>  4:    17   32
+#>  5:    17   40
+#>  6:    17   50
+#>  7:    18   42
+#>  8:    18   56
+#>  9:    18   76
+#> 10:    18   84
+#> 11:    19   36
+#> 12:    19   46
+#> 13:    19   68
+#> 14:    20   32
+#> 15:    20   48
+#> 16:    20   52
+#> 17:    20   56
+#> 18:    20   64
+#> 19:    22   66
+#> 20:    23   54
+#> 21:    24   70
+#> 22:    24   92
+#> 23:    24   93
+#> 24:    24  120
+#> 25:    25   85
+#>     speed dist
+get_chunk(cars.df, 1, keep = "speed")
+#>     speed
+#>  1:     4
+#>  2:     4
+#>  3:     7
+#>  4:     7
+#>  5:     8
+#>  6:     9
+#>  7:    10
+#>  8:    10
+#>  9:    10
+#> 10:    11
+#> 11:    11
+#> 12:    12
+#> 13:    12
+#> 14:    12
+#> 15:    12
+#> 16:    13
+#> 17:    13
+#> 18:    13
+#> 19:    13
+#> 20:    14
+#> 21:    14
+#> 22:    14
+#> 23:    14
+#> 24:    15
+#> 25:    15
+#>     speed
+
+# if full.names = TRUE then the full path to the chunk need to be provided
+get_chunk(cars.df, file.path(attr(cars.df, "path"), "1.fst"), full.names = TRUE)
+#>     speed dist
+#>  1:     4    2
+#>  2:     4   10
+#>  3:     7    4
+#>  4:     7   22
+#>  5:     8   16
+#>  6:     9   10
+#>  7:    10   18
+#>  8:    10   26
+#>  9:    10   34
+#> 10:    11   17
+#> 11:    11   28
+#> 12:    12   14
+#> 13:    12   20
+#> 14:    12   24
+#> 15:    12   28
+#> 16:    13   26
+#> 17:    13   34
+#> 18:    13   34
+#> 19:    13   46
+#> 20:    14   26
+#> 21:    14   36
+#> 22:    14   60
+#> 23:    14   80
+#> 24:    15   20
+#> 25:    15   26
+#>     speed dist
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/get_chunk_ids.html b/docs/reference/get_chunk_ids.html index b6b80fac..d54c8f15 100644 --- a/docs/reference/get_chunk_ids.html +++ b/docs/reference/get_chunk_ids.html @@ -1,67 +1,12 @@ - - - - - - - -Get the chunk IDs and files names — get_chunk_ids • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Get the chunk IDs and files names — get_chunk_ids • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Get the chunk IDs and files names

    - Source: R/get_chunk_ids.r + Source: R/get_chunk_ids.r
    @@ -165,75 +95,68 @@

    Get the chunk IDs and files names

    Get the chunk IDs and files names

    - 
    get_chunk_ids(df, ..., full.names = FALSE, strip_extension = TRUE)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    df

    a disk.frame

    ...

    passed to list.files

    full.names

    If TRUE returns the full path to the file, Defaults to FALSE

    strip_extension

    If TRUE then the file extension in the chunk_id is removed. Defaults to TRUE

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-
-# return the integer-string chunk IDs
-get_chunk_ids(cars.df)
-
    #> [1] "1" "2" "3" "4" "5" "6"
    
-# return the file name chunk IDs
-get_chunk_ids(cars.df, full.names = TRUE)
-
    #> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/1.fst"
-#> [2] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/2.fst"
-#> [3] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/3.fst"
-#> [4] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/4.fst"
-#> [5] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/5.fst"
-#> [6] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2PGAkP\\file4b14668a2842.df/6.fst"
    
-# return the file name chunk IDs with file extension
-get_chunk_ids(cars.df, strip_extension = FALSE)
-
    #> [1] "1.fst" "2.fst" "3.fst" "4.fst" "5.fst" "6.fst"
    
-# clean up cars.df
-delete(cars.df)
-
    +
    + 
    get_chunk_ids(df, ..., full.names = FALSE, strip_extension = TRUE)
    +
    + +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    ...
    +

    passed to list.files

    +
    full.names
    +

    If TRUE returns the full path to the file, Defaults to FALSE

    +
    strip_extension
    +

    If TRUE then the file extension in the chunk_id is removed. Defaults to TRUE

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# return the integer-string chunk IDs
+get_chunk_ids(cars.df)
+#> [1] "1" "2" "3" "4" "5" "6"
+
+# return the file name chunk IDs
+get_chunk_ids(cars.df, full.names = TRUE)
+#> [1] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2rQjw5\\file56f46178545a.df/1.fst"
+#> [2] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2rQjw5\\file56f46178545a.df/2.fst"
+#> [3] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2rQjw5\\file56f46178545a.df/3.fst"
+#> [4] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2rQjw5\\file56f46178545a.df/4.fst"
+#> [5] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2rQjw5\\file56f46178545a.df/5.fst"
+#> [6] "C:\\Users\\RTX2080\\AppData\\Local\\Temp\\Rtmp2rQjw5\\file56f46178545a.df/6.fst"
+
+# return the file name chunk IDs with file extension
+get_chunk_ids(cars.df, strip_extension = FALSE)
+#> [1] "1.fst" "2.fst" "3.fst" "4.fst" "5.fst" "6.fst"
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/group_by.html b/docs/reference/group_by.html index f4e80c6e..a7d39c8e 100644 --- a/docs/reference/group_by.html +++ b/docs/reference/group_by.html @@ -1,70 +1,15 @@ - - - - - - - -A function to parse the summarize function — summarise.grouped_disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -A function to parse the summarize function — summarise.grouped_disk.frame • disk.frame - - - - - - - - - - + + - - - -
    -
    - -
    - -
    +

    A function to parse the summarize function

    - Source: R/one-stage-verbs.R + Source: R/one-stage-verbs.R
    @@ -171,71 +101,60 @@

    A function to parse the summarize function

    reorganizes the chunks by the shard key.

    - 
    # S3 method for grouped_disk.frame
-summarise(.data, ...)
-
-# S3 method for grouped_disk.frame
-summarize(.data, ...)
-
-# S3 method for disk.frame
-group_by(.data, ..., add = FALSE, .drop = dplyr::group_by_drop_default(.data))
-
-# S3 method for disk.frame
-summarize(.data, ...)
-
-# S3 method for disk.frame
-summarise(.data, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    .data

    a disk.frame

    ...

    same as the dplyr::group_by

    add

    from dplyr

    .drop

    from dplyr

    - -

    See also

    - -

    hard_group_by

    +
    + 
    # S3 method for grouped_disk.frame
+summarise(.data, ...)
+
+# S3 method for grouped_disk.frame
+summarize(.data, ...)
+
+# S3 method for disk.frame
+group_by(.data, ..., add = FALSE, .drop = dplyr::group_by_drop_default(.data))
+
+# S3 method for disk.frame
+summarize(.data, ...)
+
+# S3 method for disk.frame
+summarise(.data, ...)
    +
    + +
    +

    Arguments

    +
    .data
    +

    a disk.frame

    +
    ...
    +

    same as the dplyr::group_by

    +
    add
    +

    from dplyr

    +
    .drop
    +

    from dplyr

    +
    +
    +

    See also

    +

    hard_group_by

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/groups.disk.frame.html b/docs/reference/groups.disk.frame.html index 9c8d9c05..02be8acf 100644 --- a/docs/reference/groups.disk.frame.html +++ b/docs/reference/groups.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -The shard keys of the disk.frame — groups.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -The shard keys of the disk.frame — groups.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    The shard keys of the disk.frame

    - Source: R/dplyr_verbs.r + Source: R/dplyr_verbs.r
    @@ -165,47 +95,42 @@

    The shard keys of the disk.frame

    The shard keys of the disk.frame

    - 
    # S3 method for disk.frame
-groups(x)
    - -

    Arguments

    - - - - - - -
    x

    a disk.frame

    - -

    Value

    +
    + 
    # S3 method for disk.frame
+groups(x)
    +
    +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    +
    +

    Value

    character

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/hard_arrange.html b/docs/reference/hard_arrange.html index d31b48ee..c880a71b 100644 --- a/docs/reference/hard_arrange.html +++ b/docs/reference/hard_arrange.html @@ -1,69 +1,14 @@ - - - - - - - -Perform a hard arrange — hard_arrange • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Perform a hard arrange — hard_arrange • disk.frame - - - - - - - - - + + - - - - -
    -
    - -
    - -
    +

    Perform a hard arrange

    - Source: R/hard_arrange.r + Source: R/hard_arrange.r
    @@ -169,120 +99,107 @@

    Perform a hard arrange

    row that share the same `by` value will end up in the same chunk.

    - 
    hard_arrange(df, ..., add = FALSE, .drop = FALSE)
-
-# S3 method for data.frame
-hard_arrange(df, ...)
-
-# S3 method for disk.frame
-hard_arrange(
-  df,
-  ...,
-  outdir = tempfile("tmp_disk_frame_hard_arrange"),
-  nchunks = disk.frame::nchunks(df),
-  overwrite = TRUE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df

    a disk.frame

    ...

    grouping variables

    add

    same as dplyr::arrange

    .drop

    same as dplyr::arrange

    outdir

    the output directory

    nchunks

    The number of chunks in the output. Defaults = nchunks.disk.frame(df)

    overwrite

    overwrite the out put directory

    +
    + 
    hard_arrange(df, ..., add = FALSE, .drop = FALSE)
 
+# S3 method for data.frame
+hard_arrange(df, ...)
 
-    
    Examples
    
-    
    iris.df = as.disk.frame(iris, nchunks = 2)
+# S3 method for disk.frame
+hard_arrange(
+  df,
+  ...,
+  outdir = tempfile("tmp_disk_frame_hard_arrange"),
+  nchunks = disk.frame::nchunks(df),
+  overwrite = TRUE
+)
    +
    -# arrange iris.df by specifies and ensure rows with the same specifies are in the same chunk -iris_hard.df = hard_arrange(iris.df, Species) -
    #> Appending disk.frames:
    -get_chunk(iris_hard.df, 1) -
    #> # A tibble: 50 x 5 -#> # Groups: Species [1] -#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species -#> <dbl> <dbl> <dbl> <dbl> <fct> -#> 1 6.3 3.3 6 2.5 virginica -#> 2 5.8 2.7 5.1 1.9 virginica -#> 3 7.1 3 5.9 2.1 virginica -#> 4 6.3 2.9 5.6 1.8 virginica -#> 5 6.5 3 5.8 2.2 virginica -#> 6 7.6 3 6.6 2.1 virginica -#> 7 4.9 2.5 4.5 1.7 virginica -#> 8 7.3 2.9 6.3 1.8 virginica -#> 9 6.7 2.5 5.8 1.8 virginica -#> 10 7.2 3.6 6.1 2.5 virginica -#> # ... with 40 more rows
    get_chunk(iris_hard.df, 2) -
    #> # A tibble: 50 x 5 -#> # Groups: Species [1] -#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species -#> <dbl> <dbl> <dbl> <dbl> <fct> -#> 1 6.3 3.3 6 2.5 virginica -#> 2 5.8 2.7 5.1 1.9 virginica -#> 3 7.1 3 5.9 2.1 virginica -#> 4 6.3 2.9 5.6 1.8 virginica -#> 5 6.5 3 5.8 2.2 virginica -#> 6 7.6 3 6.6 2.1 virginica -#> 7 4.9 2.5 4.5 1.7 virginica -#> 8 7.3 2.9 6.3 1.8 virginica -#> 9 6.7 2.5 5.8 1.8 virginica -#> 10 7.2 3.6 6.1 2.5 virginica -#> # ... with 40 more rows
    -# clean up cars.df -delete(iris.df) -delete(iris_hard.df) -
    +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    ...
    +

    grouping variables

    +
    add
    +

    same as dplyr::arrange

    +
    .drop
    +

    same as dplyr::arrange

    +
    outdir
    +

    the output directory

    +
    nchunks
    +

    The number of chunks in the output. Defaults = nchunks.disk.frame(df)

    +
    overwrite
    +

    overwrite the out put directory

    +
    + +
    +

    Examples

    + 
    iris.df = as.disk.frame(iris, nchunks = 2)
+
+# arrange iris.df by specifies and ensure rows with the same specifies are in the same chunk
+iris_hard.df = hard_arrange(iris.df, Species)
+#> Appending disk.frames: 
+
+get_chunk(iris_hard.df, 1)
+#> # A tibble: 50 x 5
+#> # Groups:   Species [1]
+#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species  
+#>           <dbl>       <dbl>        <dbl>       <dbl> <fct>    
+#>  1          6.3         3.3          6           2.5 virginica
+#>  2          5.8         2.7          5.1         1.9 virginica
+#>  3          7.1         3            5.9         2.1 virginica
+#>  4          6.3         2.9          5.6         1.8 virginica
+#>  5          6.5         3            5.8         2.2 virginica
+#>  6          7.6         3            6.6         2.1 virginica
+#>  7          4.9         2.5          4.5         1.7 virginica
+#>  8          7.3         2.9          6.3         1.8 virginica
+#>  9          6.7         2.5          5.8         1.8 virginica
+#> 10          7.2         3.6          6.1         2.5 virginica
+#> # ... with 40 more rows
+get_chunk(iris_hard.df, 2)
+#> # A tibble: 50 x 5
+#> # Groups:   Species [1]
+#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species  
+#>           <dbl>       <dbl>        <dbl>       <dbl> <fct>    
+#>  1          6.3         3.3          6           2.5 virginica
+#>  2          5.8         2.7          5.1         1.9 virginica
+#>  3          7.1         3            5.9         2.1 virginica
+#>  4          6.3         2.9          5.6         1.8 virginica
+#>  5          6.5         3            5.8         2.2 virginica
+#>  6          7.6         3            6.6         2.1 virginica
+#>  7          4.9         2.5          4.5         1.7 virginica
+#>  8          7.3         2.9          6.3         1.8 virginica
+#>  9          6.7         2.5          5.8         1.8 virginica
+#> 10          7.2         3.6          6.1         2.5 virginica
+#> # ... with 40 more rows
+
+# clean up cars.df
+delete(iris.df)
+delete(iris_hard.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/hard_group_by.html b/docs/reference/hard_group_by.html index 7df1019d..8dd5247b 100644 --- a/docs/reference/hard_group_by.html +++ b/docs/reference/hard_group_by.html @@ -1,69 +1,14 @@ - - - - - - - -Perform a hard group — hard_group_by • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Perform a hard group — hard_group_by • disk.frame - - - - - - - - - - + + - - - -
    -
    - -
    - -
    +

    Perform a hard group

    - Source: R/hard_group_by.r + Source: R/hard_group_by.r
    @@ -169,126 +99,108 @@

    Perform a hard group

    row that share the same `by` value will end up in the same chunk.

    - 
    hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
-
-# S3 method for data.frame
-hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
-
-# S3 method for disk.frame
-hard_group_by(
-  df,
-  ...,
-  outdir = tempfile("tmp_disk_frame_hard_group_by"),
-  nchunks = disk.frame::nchunks(df),
-  overwrite = TRUE,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL,
-  sort_split_sample_size = 100
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df

    a disk.frame

    ...

    grouping variables

    .add

    same as dplyr::group_by

    .drop

    same as dplyr::group_by

    outdir

    the output directory

    nchunks

    The number of chunks in the output. Defaults = nchunks.disk.frame(df)

    overwrite

    overwrite the out put directory

    shardby_function

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    sort_splits

    for the "sort" shardby function, a dataframe with the split values.

    desc_vars

    for the "sort" shardby function, the variables to sort descending.

    sort_split_sample_size

    for the "sort" shardby function, if sort_splits is null, the number of rows to sample per chunk for random splits.

    - - -

    Examples

    - 
    iris.df = as.disk.frame(iris, nchunks = 2)
+    
    
+    
    hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
+
+# S3 method for data.frame
+hard_group_by(df, ..., .add = FALSE, .drop = FALSE)
+
+# S3 method for disk.frame
+hard_group_by(
+  df,
+  ...,
+  outdir = tempfile("tmp_disk_frame_hard_group_by"),
+  nchunks = disk.frame::nchunks(df),
+  overwrite = TRUE,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL,
+  sort_split_sample_size = 100
+)
    
+    
    
 
-# group_by iris.df by specifies and ensure rows with the same specifies are in the same chunk
-iris_hard.df = hard_group_by(iris.df, Species)
-
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    
-get_chunk(iris_hard.df, 1)
-
    #> # A tibble: 150 x 5
-#> # Groups:   Species [3]
-#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
-#>           <dbl>       <dbl>        <dbl>       <dbl> <fct>  
-#>  1          5.1         3.5          1.4         0.2 setosa 
-#>  2          4.9         3            1.4         0.2 setosa 
-#>  3          4.7         3.2          1.3         0.2 setosa 
-#>  4          4.6         3.1          1.5         0.2 setosa 
-#>  5          5           3.6          1.4         0.2 setosa 
-#>  6          5.4         3.9          1.7         0.4 setosa 
-#>  7          4.6         3.4          1.4         0.3 setosa 
-#>  8          5           3.4          1.5         0.2 setosa 
-#>  9          4.4         2.9          1.4         0.2 setosa 
-#> 10          4.9         3.1          1.5         0.1 setosa 
-#> # ... with 140 more rows
    get_chunk(iris_hard.df, 2)
-
    #> Warning: The chunk NA does not exist; returning an empty data.table
    #> Null data.table (0 rows and 0 cols)
    
-# clean up cars.df
-delete(iris.df)
-delete(iris_hard.df)
-
    +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    ...
    +

    grouping variables

    +
    .add
    +

    same as dplyr::group_by

    +
    .drop
    +

    same as dplyr::group_by

    +
    outdir
    +

    the output directory

    +
    nchunks
    +

    The number of chunks in the output. Defaults = nchunks.disk.frame(df)

    +
    overwrite
    +

    overwrite the out put directory

    +
    shardby_function
    +

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    +
    sort_splits
    +

    for the "sort" shardby function, a dataframe with the split values.

    +
    desc_vars
    +

    for the "sort" shardby function, the variables to sort descending.

    +
    sort_split_sample_size
    +

    for the "sort" shardby function, if sort_splits is null, the number of rows to sample per chunk for random splits.

    +
    + +
    +

    Examples

    + 
    iris.df = as.disk.frame(iris, nchunks = 2)
+
+# group_by iris.df by specifies and ensure rows with the same specifies are in the same chunk
+iris_hard.df = hard_group_by(iris.df, Species)
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+
+get_chunk(iris_hard.df, 1)
+#> # A tibble: 150 x 5
+#> # Groups:   Species [3]
+#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species
+#>           <dbl>       <dbl>        <dbl>       <dbl> <fct>  
+#>  1          5.1         3.5          1.4         0.2 setosa 
+#>  2          4.9         3            1.4         0.2 setosa 
+#>  3          4.7         3.2          1.3         0.2 setosa 
+#>  4          4.6         3.1          1.5         0.2 setosa 
+#>  5          5           3.6          1.4         0.2 setosa 
+#>  6          5.4         3.9          1.7         0.4 setosa 
+#>  7          4.6         3.4          1.4         0.3 setosa 
+#>  8          5           3.4          1.5         0.2 setosa 
+#>  9          4.4         2.9          1.4         0.2 setosa 
+#> 10          4.9         3.1          1.5         0.1 setosa 
+#> # ... with 140 more rows
+get_chunk(iris_hard.df, 2)
+#> Warning: The chunk NA does not exist; returning an empty data.table
+#> Null data.table (0 rows and 0 cols)
+
+# clean up cars.df
+delete(iris.df)
+delete(iris_hard.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/head_tail.html b/docs/reference/head_tail.html index 74fe55c0..6312a2b2 100644 --- a/docs/reference/head_tail.html +++ b/docs/reference/head_tail.html @@ -1,67 +1,12 @@ - - - - - - - -Head and tail of the disk.frame — head.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Head and tail of the disk.frame — head.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Head and tail of the disk.frame

    - Source: R/disk.frame.r + Source: R/disk.frame.r
    @@ -165,74 +95,68 @@

    Head and tail of the disk.frame

    Head and tail of the disk.frame

    - 
    # S3 method for disk.frame
-head(x, n = 6L, ...)
-
-# S3 method for disk.frame
-tail(x, n = 6L, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - -
    x

    a disk.frame

    n

    number of rows to include

    ...

    passed to base::head or base::tail

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-head(cars.df)
-
    #>    speed dist
-#> 1:     4    2
-#> 2:     4   10
-#> 3:     7    4
-#> 4:     7   22
-#> 5:     8   16
-#> 6:     9   10
    tail(cars.df)
-
    #>    speed dist
-#> 1:    24   70
-#> 2:    24   92
-#> 3:    24   93
-#> 4:    24  120
-#> 5:    25   85
    
-# clean up 
-delete(cars.df)
-
    +
    + 
    # S3 method for disk.frame
+head(x, n = 6L, ...)
+
+# S3 method for disk.frame
+tail(x, n = 6L, ...)
    +
    + +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    n
    +

    number of rows to include

    +
    ...
    +

    passed to base::head or base::tail

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+head(cars.df)
+#>    speed dist
+#> 1:     4    2
+#> 2:     4   10
+#> 3:     7    4
+#> 4:     7   22
+#> 5:     8   16
+#> 6:     9   10
+tail(cars.df)
+#>    speed dist
+#> 1:    24   70
+#> 2:    24   92
+#> 3:    24   93
+#> 4:    24  120
+#> 5:    25   85
+
+# clean up 
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/index.html b/docs/reference/index.html index 4f823595..482f042c 100644 --- a/docs/reference/index.html +++ b/docs/reference/index.html @@ -1,66 +1,12 @@ - - - - - - - -Function reference • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Function reference • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Reference

    - - - - - - - - - - -
    -

    All functions

    + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    +

    All functions

    +

    add_chunk()

    Add a chunk to the disk.frame

    +

    as.data.frame(<disk.frame>)

    Convert disk.frame to data.frame by collecting all chunks

    +

    as.data.table(<disk.frame>)

    Convert disk.frame to data.table by collecting all chunks

    +

    as.disk.frame()

    Make a data.frame into a disk.frame

    +

    chunk_summarize() chunk_summarise() chunk_group_by() chunk_ungroup()

    Group by within each disk.frame

    +

    cmap() cmap_dfr() cimap() cimap_dfr() lazy() delayed() chunk_lapply() map() imap_dfr() imap() map_dfr(<disk.frame>) map_dfr(<default>)

    Apply the same function to all chunks

    +

    cmap2() map2() map_by_chunk_id()

    `cmap2` a function to two disk.frames

    +

    collect(<disk.frame>) collect_list() collect(<summarized_disk.frame>)

    Bring the disk.frame into R

    +

    colnames() names(<disk.frame>)

    Return the column names of the disk.frame

    +

    compute(<disk.frame>)

    Compute without writing

    +

    create_chunk_mapper()

    Create function that applies to each chunk if disk.frame

    +

    csv_to_disk.frame()

    Convert CSV file(s) to disk.frame format

    +

    delete()

    Delete a disk.frame

    +

    dfglm()

    Fit generalized linear models (glm) with disk.frame

    +

    df_ram_size()

    Get the size of RAM in gigabytes

    +

    disk.frame()

    Create a disk.frame from a folder

    +

    select(<disk.frame>) rename(<disk.frame>) filter(<disk.frame>) mutate(<disk.frame>) transmute(<disk.frame>) arrange(<disk.frame>) chunk_arrange() add_tally.disk.frame() do(<disk.frame>) distinct(<disk.frame>) chunk_distinct() glimpse(<disk.frame>)

    The dplyr verbs implemented for disk.frame

    +

    evalparseglue()

    Helper function to evalparse some `glue::glue` string

    +

    foverlaps.disk.frame()

    Apply data.table's foverlaps to the disk.frame

    +

    gen_datatable_synthetic()

    Generate synthetic dataset for testing

    +

    get_chunk()

    Obtain one chunk by chunk id

    +

    get_chunk_ids()

    Get the chunk IDs and files names

    +

    groups(<disk.frame>)

    The shard keys of the disk.frame

    +

    summarise(<grouped_disk.frame>) summarize(<grouped_disk.frame>) group_by(<disk.frame>) summarize(<disk.frame>) summarise(<disk.frame>)

    A function to parse the summarize function

    +

    hard_arrange()

    Perform a hard arrange

    +

    hard_group_by()

    Perform a hard group

    +

    head(<disk.frame>) tail(<disk.frame>)

    Head and tail of the disk.frame

    +

    is_disk.frame()

    Checks if a folder is a disk.frame

    +

    anti_join(<disk.frame>) full_join(<disk.frame>) inner_join(<disk.frame>) left_join(<disk.frame>) semi_join(<disk.frame>)

    Performs join/merge for disk.frames

    +

    make_glm_streaming_fn()

    A streaming function for speedglm

    +

    merge(<disk.frame>)

    Merge function for disk.frames

    +

    move_to() copy_df_to()

    Move or copy a disk.frame to another location

    +

    nchunks() nchunk()

    Returns the number of chunks in a disk.frame

    +

    nrow() ncol()

    Number of rows or columns

    +

    var_df.chunk_agg.disk.frame() var_df.collected_agg.disk.frame() sd_df.chunk_agg.disk.frame() sd_df.collected_agg.disk.frame() mean_df.chunk_agg.disk.frame() mean_df.collected_agg.disk.frame() sum_df.chunk_agg.disk.frame() sum_df.collected_agg.disk.frame() min_df.chunk_agg.disk.frame() min_df.collected_agg.disk.frame() max_df.chunk_agg.disk.frame() max_df.collected_agg.disk.frame() median_df.chunk_agg.disk.frame() median_df.collected_agg.disk.frame() n_df.chunk_agg.disk.frame() n_df.collected_agg.disk.frame() length_df.chunk_agg.disk.frame() length_df.collected_agg.disk.frame() any_df.chunk_agg.disk.frame() any_df.collected_agg.disk.frame() all_df.chunk_agg.disk.frame() all_df.collected_agg.disk.frame() n_distinct_df.chunk_agg.disk.frame() n_distinct_df.collected_agg.disk.frame() quantile_df.chunk_agg.disk.frame() quantile_df.collected_agg.disk.frame() IQR_df.chunk_agg.disk.frame() IQR_df.collected_agg.disk.frame()

    One Stage function

    +

    overwrite_check()

    Check if the outdir exists or not

    +

    print(<disk.frame>)

    Print disk.frame

    +

    pull(<disk.frame>)

    Pull a column from table similar to `dplyr::pull`.

    +

    rbindlist.disk.frame()

    rbindlist disk.frames together

    +

    rechunk()

    Increase or decrease the number of chunks in the disk.frame

    +

    recommend_nchunks()

    Recommend number of chunks based on input size

    +

    remove_chunk()

    Removes a chunk from the disk.frame

    +

    sample_frac(<disk.frame>)

    Sample n rows from a disk.frame

    +

    setup_disk.frame()

    Set up disk.frame environment

    +

    shard() distribute()

    Shard a data.frame/data.table or disk.frame into chunk and saves it into a disk.frame

    +

    shardkey()

    Returns the shardkey (not implemented yet)

    +

    shardkey_equal()

    Compare two disk.frame shardkeys

    +

    show_ceremony() ceremony_text() show_boilerplate() insert_ceremony()

    Show the code to setup disk.frame

    +

    srckeep() srckeepchunks()

    Keep only the variables from the input listed in selections

    +

    `[`(<disk.frame>)

    [ interface for disk.frame using fst backend

    +

    tbl_vars(<disk.frame>) group_vars(<disk.frame>)

    Column names for RStudio auto-complete

    +

    write_disk.frame() output_disk.frame()

    Write disk.frame to disk

    +

    zip_to_disk.frame()

    `zip_to_disk.frame` is used to read and convert every CSV file within the zip file to disk.frame format

    - +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/is_disk.frame.html b/docs/reference/is_disk.frame.html index b64b33ea..c56ab38f 100644 --- a/docs/reference/is_disk.frame.html +++ b/docs/reference/is_disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Checks if a folder is a disk.frame — is_disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Checks if a folder is a disk.frame — is_disk.frame • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Checks if a folder is a disk.frame

    - Source: R/is_disk.frame.r + Source: R/is_disk.frame.r
    @@ -165,52 +95,50 @@

    Checks if a folder is a disk.frame

    Checks if a folder is a disk.frame

    - 
    is_disk.frame(df)
    - -

    Arguments

    - - - - - - -
    df

    a disk.frame or directory to check

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
+    
    
+    
    is_disk.frame(df)
    
+    
    
 
-is_disk.frame(cars) # FALSE
-
    #> [1] FALSE
    is_disk.frame(cars.df) # TRUE
-
    #> [1] TRUE
    
-# clean up cars.df
-delete(cars.df)
-
    +
    +

    Arguments

    +
    df
    +

    a disk.frame or directory to check

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+is_disk.frame(cars) # FALSE
+#> [1] FALSE
+is_disk.frame(cars.df) # TRUE
+#> [1] TRUE
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/join.html b/docs/reference/join.html index 3abfbc81..6bdc7ff7 100644 --- a/docs/reference/join.html +++ b/docs/reference/join.html @@ -1,67 +1,12 @@ - - - - - - - -Performs join/merge for disk.frames — anti_join.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Performs join/merge for disk.frames — anti_join.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Performs join/merge for disk.frames

    - Source: R/anti_join.r, R/full_join.r, R/inner_join.r, and 2 more + Source: R/anti_join.r, R/full_join.r, R/inner_join.r, and 2 more
    @@ -165,185 +95,208 @@

    Performs join/merge for disk.frames

    Performs join/merge for disk.frames

    - 
    # S3 method for disk.frame
-anti_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_anti_join"),
-  merge_by_chunk_id = FALSE,
-  overwrite = TRUE,
-  .progress = FALSE
-)
-
-# S3 method for disk.frame
-full_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_full_join"),
-  overwrite = TRUE,
-  merge_by_chunk_id,
-  .progress = FALSE
-)
-
-# S3 method for disk.frame
-inner_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_inner_join"),
-  merge_by_chunk_id = NULL,
-  overwrite = TRUE,
-  .progress = FALSE
-)
-
-# S3 method for disk.frame
-left_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_left_join"),
-  merge_by_chunk_id = FALSE,
-  overwrite = TRUE,
-  .progress = FALSE
-)
-
-# S3 method for disk.frame
-semi_join(
-  x,
-  y,
-  by = NULL,
-  copy = FALSE,
-  ...,
-  outdir = tempfile("tmp_disk_frame_semi_join"),
-  merge_by_chunk_id = FALSE,
-  overwrite = TRUE,
-  .progress = FALSE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    x

    a disk.frame

    y

    a data.frame or disk.frame. If data.frame then returns lazily; if disk.frame it performs the join eagerly and return a disk.frame

    by

    join by

    copy

    same as dplyr::anti_join

    ...

    same as dplyr's joins

    outdir

    output directory for disk.frame

    merge_by_chunk_id

    the merge is performed by chunk id

    overwrite

    overwrite output directory

    .progress

    Show progress or not. Defaults to FALSE

    - -

    Value

    +
    + 
    # S3 method for disk.frame
+anti_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_anti_join"),
+  merge_by_chunk_id = FALSE,
+  overwrite = TRUE,
+  .progress = FALSE
+)
+
+# S3 method for disk.frame
+full_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_full_join"),
+  overwrite = TRUE,
+  merge_by_chunk_id,
+  .progress = FALSE
+)
+
+# S3 method for disk.frame
+inner_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_inner_join"),
+  merge_by_chunk_id = NULL,
+  overwrite = TRUE,
+  .progress = FALSE
+)
+
+# S3 method for disk.frame
+left_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_left_join"),
+  merge_by_chunk_id = FALSE,
+  overwrite = TRUE,
+  .progress = FALSE
+)
+
+# S3 method for disk.frame
+semi_join(
+  x,
+  y,
+  by = NULL,
+  copy = FALSE,
+  ...,
+  outdir = tempfile("tmp_disk_frame_semi_join"),
+  merge_by_chunk_id = FALSE,
+  overwrite = TRUE,
+  .progress = FALSE
+)
    +
    +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    y
    +

    a data.frame or disk.frame. If data.frame then returns lazily; if disk.frame it performs the join eagerly and return a disk.frame

    +
    by
    +

    join by

    +
    copy
    +

    same as dplyr::anti_join

    +
    ...
    +

    same as dplyr's joins

    +
    outdir
    +

    output directory for disk.frame

    +
    merge_by_chunk_id
    +

    the merge is performed by chunk id

    +
    overwrite
    +

    overwrite output directory

    +
    .progress
    +

    Show progress or not. Defaults to FALSE

    +
    +
    +

    Value

    disk.frame or data.frame/data.table

    +
    -

    Examples

    - 
    df.df = as.disk.frame(data.frame(x = 1:3, y = 4:6), overwrite = TRUE)
-df2.df = as.disk.frame(data.frame(x = 1:2, z = 10:11), overwrite = TRUE)
-
-anti_joined.df = anti_join(df.df, df2.df) 
-
    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    
-anti_joined.df %>% collect
-
    #>    x y
-#> 1: 3 6
    
-anti_joined.data.frame = anti_join(df.df, data.frame(x = 1:2, z = 10:11))
-
    #> Joining, by = "x"
    #> Joining, by = "x"
    #> Joining, by = "x"
    
-# clean up
-delete(df.df)
-delete(df2.df)
-delete(anti_joined.df)
-cars.df = as.disk.frame(cars)
-
-join.df = full_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
-
-# clean up cars.df
-delete(cars.df)
-delete(join.df)
-cars.df = as.disk.frame(cars)
-
-join.df = inner_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
-
-# clean up cars.df
-delete(cars.df)
-delete(join.df)
-cars.df = as.disk.frame(cars)
-
-join.df = left_join(cars.df, cars.df)
-
    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    
-# clean up cars.df
-delete(cars.df)
-delete(join.df)
-cars.df = as.disk.frame(cars)
-
-join.df = semi_join(cars.df, cars.df)
-
    #> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    
-# clean up cars.df
-delete(cars.df)
-delete(join.df)
-
    +
    +

    Examples

    + 
    df.df = as.disk.frame(data.frame(x = 1:3, y = 4:6), overwrite = TRUE)
+df2.df = as.disk.frame(data.frame(x = 1:2, z = 10:11), overwrite = TRUE)
+
+anti_joined.df = anti_join(df.df, df2.df) 
+#> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+
+anti_joined.df %>% collect
+#>    x y
+#> 1: 3 6
+
+anti_joined.data.frame = anti_join(df.df, data.frame(x = 1:2, z = 10:11))
+#> Joining, by = "x"
+#> Joining, by = "x"
+#> Joining, by = "x"
+
+# clean up
+delete(df.df)
+delete(df2.df)
+delete(anti_joined.df)
+cars.df = as.disk.frame(cars)
+
+join.df = full_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
+
+# clean up cars.df
+delete(cars.df)
+delete(join.df)
+cars.df = as.disk.frame(cars)
+
+join.df = inner_join(cars.df, cars.df, merge_by_chunk_id = TRUE)
+
+# clean up cars.df
+delete(cars.df)
+delete(join.df)
+cars.df = as.disk.frame(cars)
+
+join.df = left_join(cars.df, cars.df)
+#> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+
+# clean up cars.df
+delete(cars.df)
+delete(join.df)
+cars.df = as.disk.frame(cars)
+
+join.df = semi_join(cars.df, cars.df)
+#> Warning: merge_by_chunk_id = FALSE. This will take significantly longer and the preparations needed are performed eagerly which may lead to poor performance. Consider making y a data.frame or set merge_by_chunk_id = TRUE for better performance.
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+
+# clean up cars.df
+delete(cars.df)
+delete(join.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/make_glm_streaming_fn.html b/docs/reference/make_glm_streaming_fn.html index 1f76192a..91b68bc6 100644 --- a/docs/reference/make_glm_streaming_fn.html +++ b/docs/reference/make_glm_streaming_fn.html @@ -1,67 +1,12 @@ - - - - - - - -A streaming function for speedglm — make_glm_streaming_fn • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -A streaming function for speedglm — make_glm_streaming_fn • disk.frame - + + - - - -
    -
    - -
    - -
    +

    A streaming function for speedglm

    - Source: R/make_glm_streaming_fn.r + Source: R/make_glm_streaming_fn.r
    @@ -165,119 +95,116 @@

    A streaming function for speedglm

    Define a function that can be used to feed data into speedglm and biglm

    - 
    make_glm_streaming_fn(data, verbose = FALSE)
    - -

    Arguments

    - - - - - - - - - - -
    data

    a disk.frame

    verbose

    Whether to print the status of data loading. Default to FALSE

    - -

    Value

    +
    + 
    make_glm_streaming_fn(data, verbose = FALSE)
    +
    +
    +

    Arguments

    +
    data
    +

    a disk.frame

    +
    verbose
    +

    Whether to print the status of data loading. Default to FALSE

    +
    +
    +

    Value

    return a function, fn, that can be used as the data argument in biglm::bigglm or speedglm::shglm

    -

    See also

    - -

    Other Machine Learning (ML): -dfglm()

    - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-streamacq = make_glm_streaming_fn(cars.df, verbose = FALSE)
+    
    
+    
    
+    
    See also
    
+    
    Other Machine Learning (ML): 
+dfglm()
    
+    
    
 
-majorv = as.integer(version$major)
-minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
-if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
-  m = biglm::bigglm(dist ~ speed, data = streamacq)
-  summary(m)
-  predict(m, get_chunk(cars.df, 1))
-  predict(m, collect(cars.df, 1))
-} else {
-  m = speedglm::shglm(dist ~ speed, data = streamacq)
-}
-
    #> [,1] -#> 1 -1.849460 -#> 2 -1.849460 -#> 3 9.947766 -#> 4 9.947766 -#> 5 13.880175 -#> 6 17.812584 -#> 7 21.744993 -#> 8 21.744993 -#> 9 21.744993 -#> 10 25.677401 -#> 11 25.677401 -#> 12 29.609810 -#> 13 29.609810 -#> 14 29.609810 -#> 15 29.609810 -#> 16 33.542219 -#> 17 33.542219 -#> 18 33.542219 -#> 19 33.542219 -#> 20 37.474628 -#> 21 37.474628 -#> 22 37.474628 -#> 23 37.474628 -#> 24 41.407036 -#> 25 41.407036 -#> 26 41.407036 -#> 27 45.339445 -#> 28 45.339445 -#> 29 49.271854 -#> 30 49.271854 -#> 31 49.271854 -#> 32 53.204263 -#> 33 53.204263 -#> 34 53.204263 -#> 35 53.204263 -#> 36 57.136672 -#> 37 57.136672 -#> 38 57.136672 -#> 39 61.069080 -#> 40 61.069080 -#> 41 61.069080 -#> 42 61.069080 -#> 43 61.069080 -#> 44 68.933898 -#> 45 72.866307 -#> 46 76.798715 -#> 47 76.798715 -#> 48 76.798715 -#> 49 76.798715 -#> 50 80.731124
    +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+streamacq = make_glm_streaming_fn(cars.df, verbose = FALSE)
+
+majorv = as.integer(version$major)
+minorv = as.integer(strsplit(version$minor, ".", fixed=TRUE)[[1]][1])
+if(((majorv == 3) & (minorv >= 6)) | (majorv > 3)) {
+  m = biglm::bigglm(dist ~ speed, data = streamacq)
+  summary(m)
+  predict(m, get_chunk(cars.df, 1))
+  predict(m, collect(cars.df, 1))
+} else {
+  m = speedglm::shglm(dist ~ speed, data = streamacq)
+}
+#>         [,1]
+#> 1  -1.849460
+#> 2  -1.849460
+#> 3   9.947766
+#> 4   9.947766
+#> 5  13.880175
+#> 6  17.812584
+#> 7  21.744993
+#> 8  21.744993
+#> 9  21.744993
+#> 10 25.677401
+#> 11 25.677401
+#> 12 29.609810
+#> 13 29.609810
+#> 14 29.609810
+#> 15 29.609810
+#> 16 33.542219
+#> 17 33.542219
+#> 18 33.542219
+#> 19 33.542219
+#> 20 37.474628
+#> 21 37.474628
+#> 22 37.474628
+#> 23 37.474628
+#> 24 41.407036
+#> 25 41.407036
+#> 26 41.407036
+#> 27 45.339445
+#> 28 45.339445
+#> 29 49.271854
+#> 30 49.271854
+#> 31 49.271854
+#> 32 53.204263
+#> 33 53.204263
+#> 34 53.204263
+#> 35 53.204263
+#> 36 57.136672
+#> 37 57.136672
+#> 38 57.136672
+#> 39 61.069080
+#> 40 61.069080
+#> 41 61.069080
+#> 42 61.069080
+#> 43 61.069080
+#> 44 68.933898
+#> 45 72.866307
+#> 46 76.798715
+#> 47 76.798715
+#> 48 76.798715
+#> 49 76.798715
+#> 50 80.731124
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/merge.disk.frame.html b/docs/reference/merge.disk.frame.html index b2d35726..26cafba5 100644 --- a/docs/reference/merge.disk.frame.html +++ b/docs/reference/merge.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Merge function for disk.frames — merge.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Merge function for disk.frames — merge.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Merge function for disk.frames

    - Source: R/merge.disk.frame.r + Source: R/merge.disk.frame.r
    @@ -165,86 +95,70 @@

    Merge function for disk.frames

    Merge function for disk.frames

    - 
    # S3 method for disk.frame
-merge(
-  x,
-  y,
-  by,
-  outdir = tempfile(fileext = ".df"),
-  ...,
-  merge_by_chunk_id = FALSE,
-  overwrite = FALSE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    x

    a disk.frame

    y

    a disk.frame or data.frame

    by

    the merge by keys

    outdir

    The output directory for the disk.frame

    ...

    passed to merge and cmap.disk.frame

    merge_by_chunk_id

    if TRUE then only chunks in df1 and df2 with the same chunk id will get merged

    overwrite

    overwrite the outdir or not

    - - -

    Examples

    - 
    b = as.disk.frame(data.frame(a = 51:150, b = 1:100))
-d = as.disk.frame(data.frame(a = 151:250, b = 1:100))
-bd.df = merge(b, d, by = "b", merge_by_chunk_id = TRUE)
-
-# clean up cars.df
-delete(b)
-delete(d)
-delete(bd.df)
-
    +
    + 
    # S3 method for disk.frame
+merge(
+  x,
+  y,
+  by,
+  outdir = tempfile(fileext = ".df"),
+  ...,
+  merge_by_chunk_id = FALSE,
+  overwrite = FALSE
+)
    +
    + +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    y
    +

    a disk.frame or data.frame

    +
    by
    +

    the merge by keys

    +
    outdir
    +

    The output directory for the disk.frame

    +
    ...
    +

    passed to merge and cmap.disk.frame

    +
    merge_by_chunk_id
    +

    if TRUE then only chunks in df1 and df2 with the same chunk id will get merged

    +
    overwrite
    +

    overwrite the outdir or not

    +
    + +
    +

    Examples

    + 
    b = as.disk.frame(data.frame(a = 51:150, b = 1:100))
+d = as.disk.frame(data.frame(a = 151:250, b = 1:100))
+bd.df = merge(b, d, by = "b", merge_by_chunk_id = TRUE)
+
+# clean up cars.df
+delete(b)
+delete(d)
+delete(bd.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/move_to.html b/docs/reference/move_to.html index 50090a6c..9037467e 100644 --- a/docs/reference/move_to.html +++ b/docs/reference/move_to.html @@ -1,67 +1,12 @@ - - - - - - - -Move or copy a disk.frame to another location — move_to • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Move or copy a disk.frame to another location — move_to • disk.frame - + + - - - -
    -
    - -
    - -
    +

    Move or copy a disk.frame to another location

    - Source: R/move_to.r + Source: R/move_to.r
    @@ -165,71 +95,62 @@

    Move or copy a disk.frame to another location

    Move or copy a disk.frame to another location

    - 
    move_to(df, outdir, ..., copy = FALSE)
-
-copy_df_to(df, outdir, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    df

    The disk.frame

    outdir

    The new location

    ...

    NOT USED

    copy

    Merely copy and not move

    - -

    Value

    +
    + 
    move_to(df, outdir, ..., copy = FALSE)
 
-    
    a disk.frame
    
-
-    
    Examples
    
-    
    cars.df = as.disk.frame(cars)
-
-cars_copy.df = copy_df_to(cars.df, outdir = tempfile(fileext=".df"))
+copy_df_to(df, outdir, ...)
    +
    -cars2.df = move_to(cars.df, outdir = tempfile(fileext=".df")) +
    +

    Arguments

    +
    df
    +

    The disk.frame

    +
    outdir
    +

    The new location

    +
    ...
    +

    NOT USED

    +
    copy
    +

    Merely copy and not move

    +
    +
    +

    Value

    +

    a disk.frame

    +
    -# clean up -delete(cars_copy.df) -delete(cars2.df) -
    +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+cars_copy.df = copy_df_to(cars.df, outdir = tempfile(fileext=".df"))
+
+cars2.df = move_to(cars.df, outdir = tempfile(fileext=".df"))
+
+# clean up
+delete(cars_copy.df)
+delete(cars2.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/nchunks.html b/docs/reference/nchunks.html index 91ce528c..988619c1 100644 --- a/docs/reference/nchunks.html +++ b/docs/reference/nchunks.html @@ -1,67 +1,12 @@ - - - - - - - -Returns the number of chunks in a disk.frame — nchunks • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Returns the number of chunks in a disk.frame — nchunks • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Returns the number of chunks in a disk.frame

    - Source: R/nchunks.r + Source: R/nchunks.r
    @@ -165,69 +95,63 @@

    Returns the number of chunks in a disk.frame

    Returns the number of chunks in a disk.frame

    - 
    nchunks(df, ...)
-
-nchunk(df, ...)
-
-# S3 method for disk.frame
-nchunk(df, ...)
-
-# S3 method for disk.frame
-nchunks(df, skip.ready.check = FALSE, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - -
    df

    a disk.frame

    ...

    not used

    skip.ready.check

    NOT implemented

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-
-# return the number of chunks
-nchunks(cars.df)
-
    #> [1] 6
    nchunk(cars.df)
-
    #> [1] 6
    
-# clean up cars.df
-delete(cars.df)
-
    +
    + 
    nchunks(df, ...)
+
+nchunk(df, ...)
+
+# S3 method for disk.frame
+nchunk(df, ...)
+
+# S3 method for disk.frame
+nchunks(df, skip.ready.check = FALSE, ...)
    +
    + +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    ...
    +

    not used

    +
    skip.ready.check
    +

    NOT implemented

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# return the number of chunks
+nchunks(cars.df)
+#> [1] 6
+nchunk(cars.df)
+#> [1] 6
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/ncol_nrow.html b/docs/reference/ncol_nrow.html index 5d39a4fa..a8a9d09a 100644 --- a/docs/reference/ncol_nrow.html +++ b/docs/reference/ncol_nrow.html @@ -1,67 +1,12 @@ - - - - - - - -Number of rows or columns — nrow • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Number of rows or columns — nrow • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Number of rows or columns

    - Source: R/ncol-nrow.r + Source: R/ncol-nrow.r
    @@ -165,65 +95,61 @@

    Number of rows or columns

    Number of rows or columns

    - 
    nrow(df, ...)
-
-# S3 method for disk.frame
-nrow(df, ...)
-
-ncol(df)
-
-# S3 method for disk.frame
-ncol(df)
    +
    + 
    nrow(df, ...)
 
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
    df
    a disk.frame
    ...
    passed to base::nrow
    
+# S3 method for disk.frame
+nrow(df, ...)
 
+ncol(df)
 
-    
    Examples
    
-    
    cars.df = as.disk.frame(cars)
+# S3 method for disk.frame
+ncol(df)
    +
    -# return total number of column and rows -ncol(cars.df) -
    #> [1] 2
    nrow(cars.df) -
    #> [1] 50
    -# clean up cars.df -delete(cars.df) -
    +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    ...
    +

    passed to base::nrow

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# return total number of column and rows
+ncol(cars.df)
+#> [1] 2
+nrow(cars.df)
+#> [1] 50
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/one-stage-group-by-verbs.html b/docs/reference/one-stage-group-by-verbs.html index ec4a929e..22b7a93c 100644 --- a/docs/reference/one-stage-group-by-verbs.html +++ b/docs/reference/one-stage-group-by-verbs.html @@ -1,69 +1,14 @@ - - - - - - - -One Stage function — var_df.chunk_agg.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -One Stage function — var_df.chunk_agg.disk.frame • disk.frame - - - - - - - - - - - - - + + -
    -
    - -
    - -
    +

    One Stage function

    - Source: R/one-stage-verbs.R + Source: R/one-stage-verbs.R
    @@ -169,109 +99,97 @@

    One Stage function

    mean collected_agg

    - 
    var_df.chunk_agg.disk.frame(x, na.rm = FALSE)
+    
    
+    
    var_df.chunk_agg.disk.frame(x, na.rm = FALSE)
 
-var_df.collected_agg.disk.frame(listx)
+var_df.collected_agg.disk.frame(listx)
 
-sd_df.chunk_agg.disk.frame(x, na.rm = FALSE)
+sd_df.chunk_agg.disk.frame(x, na.rm = FALSE)
 
-sd_df.collected_agg.disk.frame(listx)
+sd_df.collected_agg.disk.frame(listx)
 
-mean_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
+mean_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
 
-mean_df.collected_agg.disk.frame(listx)
+mean_df.collected_agg.disk.frame(listx)
 
-sum_df.chunk_agg.disk.frame(x, ...)
+sum_df.chunk_agg.disk.frame(x, ...)
 
-sum_df.collected_agg.disk.frame(listx, ...)
+sum_df.collected_agg.disk.frame(listx, ...)
 
-min_df.chunk_agg.disk.frame(x, ...)
+min_df.chunk_agg.disk.frame(x, ...)
 
-min_df.collected_agg.disk.frame(listx, ...)
+min_df.collected_agg.disk.frame(listx, ...)
 
-max_df.chunk_agg.disk.frame(x, ...)
+max_df.chunk_agg.disk.frame(x, ...)
 
-max_df.collected_agg.disk.frame(listx, ...)
+max_df.collected_agg.disk.frame(listx, ...)
 
-median_df.chunk_agg.disk.frame(x, ...)
+median_df.chunk_agg.disk.frame(x, ...)
 
-median_df.collected_agg.disk.frame(listx, ...)
+median_df.collected_agg.disk.frame(listx, ...)
 
-n_df.chunk_agg.disk.frame(...)
+n_df.chunk_agg.disk.frame(...)
 
-n_df.collected_agg.disk.frame(listx, ...)
+n_df.collected_agg.disk.frame(listx, ...)
 
-length_df.chunk_agg.disk.frame(x, ...)
+length_df.chunk_agg.disk.frame(x, ...)
 
-length_df.collected_agg.disk.frame(listx, ...)
+length_df.collected_agg.disk.frame(listx, ...)
 
-any_df.chunk_agg.disk.frame(x, ...)
+any_df.chunk_agg.disk.frame(x, ...)
 
-any_df.collected_agg.disk.frame(listx, ...)
+any_df.collected_agg.disk.frame(listx, ...)
 
-all_df.chunk_agg.disk.frame(x, ...)
+all_df.chunk_agg.disk.frame(x, ...)
 
-all_df.collected_agg.disk.frame(listx, ...)
+all_df.collected_agg.disk.frame(listx, ...)
 
-n_distinct_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
+n_distinct_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
 
-n_distinct_df.collected_agg.disk.frame(listx, ...)
+n_distinct_df.collected_agg.disk.frame(listx, ...)
 
-quantile_df.chunk_agg.disk.frame(x, ...)
+quantile_df.chunk_agg.disk.frame(x, ...)
 
-quantile_df.collected_agg.disk.frame(listx, ...)
+quantile_df.collected_agg.disk.frame(listx, ...)
 
-IQR_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
+IQR_df.chunk_agg.disk.frame(x, na.rm = FALSE, ...)
 
-IQR_df.collected_agg.disk.frame(listx, ...)
    
-
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
    x
    the input
    na.rm
    Remove NAs. TRUE of FALSE
    listx
    a list
    ...
    additional options
    
+IQR_df.collected_agg.disk.frame(listx, ...)
    +
    +
    +

    Arguments

    +
    x
    +

    the input

    +
    na.rm
    +

    Remove NAs. TRUE of FALSE

    +
    listx
    +

    a list

    +
    ...
    +

    additional options

    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/overwrite_check.html b/docs/reference/overwrite_check.html index 386c7661..f4d07be2 100644 --- a/docs/reference/overwrite_check.html +++ b/docs/reference/overwrite_check.html @@ -1,67 +1,12 @@ - - - - - - - -Check if the outdir exists or not — overwrite_check • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Check if the outdir exists or not — overwrite_check • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Check if the outdir exists or not

    - Source: R/overwrite_check.r + Source: R/overwrite_check.r
    @@ -165,55 +95,49 @@

    Check if the outdir exists or not

    If the overwrite is TRUE then the folder will be deleted, otherwise the folder will be created.

    - 
    overwrite_check(outdir, overwrite)
    - -

    Arguments

    - - - - - - - - - - -
    outdir

    the output directory

    overwrite

    TRUE or FALSE if `outdir`` exists and overwrite = FALSE then throw an error

    - - -

    Examples

    - 
    tf = tempfile()
-overwrite_check(tf, overwrite = FALSE)
-overwrite_check(tf, overwrite = TRUE)
-
-# clean up
-fs::dir_delete(tf)
-
    +
    + 
    overwrite_check(outdir, overwrite)
    +
    + +
    +

    Arguments

    +
    outdir
    +

    the output directory

    +
    overwrite
    +

    TRUE or FALSE if `outdir`` exists and overwrite = FALSE then throw an error

    +
    + +
    +

    Examples

    + 
    tf = tempfile()
+overwrite_check(tf, overwrite = FALSE)
+overwrite_check(tf, overwrite = TRUE)
+
+# clean up
+fs::dir_delete(tf)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/print.disk.frame.html b/docs/reference/print.disk.frame.html index b3752bd0..a0ef8db4 100644 --- a/docs/reference/print.disk.frame.html +++ b/docs/reference/print.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Print disk.frame — print.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Print disk.frame — print.disk.frame • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Print disk.frame

    - Source: R/print.disk.frame.r + Source: R/print.disk.frame.r
    @@ -165,48 +95,40 @@

    Print disk.frame

    a new print method for disk.frame

    - 
    # S3 method for disk.frame
-print(x, ...)
    - -

    Arguments

    - - - - - - - - - - -
    x

    disk.frame

    ...

    not used

    +
    + 
    # S3 method for disk.frame
+print(x, ...)
    +
    +
    +

    Arguments

    +
    x
    +

    disk.frame

    +
    ...
    +

    not used

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/pull.disk.frame.html b/docs/reference/pull.disk.frame.html index 4bcc7b8a..ff438239 100644 --- a/docs/reference/pull.disk.frame.html +++ b/docs/reference/pull.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Pull a column from table similar to `dplyr::pull`. — pull.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Pull a column from table similar to `dplyr::pull`. — pull.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Pull a column from table similar to `dplyr::pull`.

    - Source: R/pull.r + Source: R/pull.r
    @@ -165,56 +95,44 @@

    Pull a column from table similar to `dplyr::pull`.

    Pull a column from table similar to `dplyr::pull`.

    - 
    # S3 method for disk.frame
-pull(.data, var = -1, name = NULL, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    .data

    The disk.frame

    var

    can be an positive or negative integer or a character/string. See dplyr::pull documentation

    name

    See dplyr::pull documentation

    ...

    Not used, kept for compatibility with `dplyr::pull`

    +
    + 
    # S3 method for disk.frame
+pull(.data, var = -1, name = NULL, ...)
    +
    +
    +

    Arguments

    +
    .data
    +

    The disk.frame

    +
    var
    +

    can be an positive or negative integer or a character/string. See dplyr::pull documentation

    +
    name
    +

    See dplyr::pull documentation

    +
    ...
    +

    Not used, kept for compatibility with `dplyr::pull`

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/rbindlist.disk.frame.html b/docs/reference/rbindlist.disk.frame.html index 2ea44cfb..081dadec 100644 --- a/docs/reference/rbindlist.disk.frame.html +++ b/docs/reference/rbindlist.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -rbindlist disk.frames together — rbindlist.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -rbindlist disk.frames together — rbindlist.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    rbindlist disk.frames together

    - Source: R/rbindlist.disk.frame.r + Source: R/rbindlist.disk.frame.r
    @@ -165,85 +95,70 @@

    rbindlist disk.frames together

    rbindlist disk.frames together

    - 
    rbindlist.disk.frame(
-  df_list,
-  outdir = tempfile(fileext = ".df"),
-  by_chunk_id = TRUE,
-  parallel = TRUE,
-  compress = 50,
-  overwrite = TRUE,
-  .progress = TRUE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df_list

    A list of disk.frames

    outdir

    Output directory of the row-bound disk.frames

    by_chunk_id

    If TRUE then only the chunks with the same chunk IDs will be bound

    parallel

    if TRUE then bind multiple disk.frame simultaneously, Defaults to TRUE

    compress

    0-100, 100 being the highest compression rate.

    overwrite

    overwrite the output directory

    .progress

    A logical, for whether or not to print a progress bar for multiprocess, multisession, and multicore plans. From furrr

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-
-# row-bind two disk.frames
-cars2.df = rbindlist.disk.frame(list(cars.df, cars.df))
-
    #> Appending disk.frames: 
    
-# clean up cars.df
-delete(cars.df)
-delete(cars2.df)
-
    +
    + 
    rbindlist.disk.frame(
+  df_list,
+  outdir = tempfile(fileext = ".df"),
+  by_chunk_id = TRUE,
+  parallel = TRUE,
+  compress = 50,
+  overwrite = TRUE,
+  .progress = TRUE
+)
    +
    + +
    +

    Arguments

    +
    df_list
    +

    A list of disk.frames

    +
    outdir
    +

    Output directory of the row-bound disk.frames

    +
    by_chunk_id
    +

    If TRUE then only the chunks with the same chunk IDs will be bound

    +
    parallel
    +

    if TRUE then bind multiple disk.frame simultaneously, Defaults to TRUE

    +
    compress
    +

    0-100, 100 being the highest compression rate.

    +
    overwrite
    +

    overwrite the output directory

    +
    .progress
    +

    A logical, for whether or not to print a progress bar for multiprocess, multisession, and multicore plans. From furrr

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# row-bind two disk.frames
+cars2.df = rbindlist.disk.frame(list(cars.df, cars.df))
+#> Appending disk.frames: 
+
+# clean up cars.df
+delete(cars.df)
+delete(cars2.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/rechunk.html b/docs/reference/rechunk.html index fb5df9b6..1ad617e3 100644 --- a/docs/reference/rechunk.html +++ b/docs/reference/rechunk.html @@ -1,67 +1,12 @@ - - - - - - - -Increase or decrease the number of chunks in the disk.frame — rechunk • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Increase or decrease the number of chunks in the disk.frame — rechunk • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Increase or decrease the number of chunks in the disk.frame

    - Source: R/rechunk.r + Source: R/rechunk.r
    @@ -165,100 +95,88 @@

    Increase or decrease the number of chunks in the disk.frame

    Increase or decrease the number of chunks in the disk.frame

    - 
    rechunk(
-  df,
-  nchunks,
-  outdir = attr(df, "path", exact = TRUE),
-  shardby = NULL,
-  overwrite = TRUE,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df

    the disk.frame to rechunk

    nchunks

    number of chunks

    outdir

    the output directory

    shardby

    the shardkeys

    overwrite

    overwrite the output directory

    shardby_function

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    sort_splits

    for the "sort" shardby function, a dataframe with the split values.

    desc_vars

    for the "sort" shardby function, the variables to sort descending.

    - - -

    Examples

    - 
    # create a disk.frame with 2 chunks in tempdir()
-cars.df = as.disk.frame(cars, nchunks = 2)
+    
    
+    
    rechunk(
+  df,
+  nchunks,
+  outdir = attr(df, "path", exact = TRUE),
+  shardby = NULL,
+  overwrite = TRUE,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL
+)
    
+    
    
 
-# re-chunking cars.df to 3 chunks, done "in-place" to the same folder as cars.df
-rechunk(cars.df, 3)
-
    #> files have been backed up to temporary dir C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\back_up_tmp_dir4b144a1a2eda. You can recover there files until you restart your R session
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1466eb2237.df"
-#> nchunks: 3
-#> nrow (at source): 50
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    
-new_path = tempfile(fileext = ".df")
-# re-chunking cars.df to 4 chunks, shard by speed, and done "out-of-place" to a new directory
-cars2.df = rechunk(cars.df, 4, outdir=new_path, shardby = "speed")
-
    #> Hashing...
    #> Hashing...
    #> Hashing...
    #> Appending disk.frames: 
    
-# clean up cars.df
-delete(cars.df)
-delete(cars2.df)
-
    +
    +

    Arguments

    +
    df
    +

    the disk.frame to rechunk

    +
    nchunks
    +

    number of chunks

    +
    outdir
    +

    the output directory

    +
    shardby
    +

    the shardkeys

    +
    overwrite
    +

    overwrite the output directory

    +
    shardby_function
    +

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    +
    sort_splits
    +

    for the "sort" shardby function, a dataframe with the split values.

    +
    desc_vars
    +

    for the "sort" shardby function, the variables to sort descending.

    +
    + +
    +

    Examples

    + 
    # create a disk.frame with 2 chunks in tempdir()
+cars.df = as.disk.frame(cars, nchunks = 2)
+
+# re-chunking cars.df to 3 chunks, done "in-place" to the same folder as cars.df
+rechunk(cars.df, 3)
+#> files have been backed up to temporary dir C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5\back_up_tmp_dir56f4356b56cb. You can recover there files until you restart your R session
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5\file56f4c8a34c9.df"
+#> nchunks: 3
+#> nrow (at source): 50
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+
+new_path = tempfile(fileext = ".df")
+# re-chunking cars.df to 4 chunks, shard by speed, and done "out-of-place" to a new directory
+cars2.df = rechunk(cars.df, 4, outdir=new_path, shardby = "speed")
+#> Hashing...
+#> Hashing...
+#> Hashing...
+#> Appending disk.frames: 
+
+# clean up cars.df
+delete(cars.df)
+delete(cars2.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/recommend_nchunks.html b/docs/reference/recommend_nchunks.html index 1b43a82d..e549d81c 100644 --- a/docs/reference/recommend_nchunks.html +++ b/docs/reference/recommend_nchunks.html @@ -1,68 +1,13 @@ - - - - - - - -Recommend number of chunks based on input size — recommend_nchunks • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Recommend number of chunks based on input size — recommend_nchunks • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Recommend number of chunks based on input size

    - Source: R/recommend_nchunks.r + Source: R/recommend_nchunks.r
    @@ -167,77 +97,67 @@

    Recommend number of chunks based on input size

    into. It can accept filesizes in bytes (as integer) or a data.frame

    - 
    recommend_nchunks(
-  df,
-  type = "csv",
-  minchunks = data.table::getDTthreads(),
-  conservatism = 8,
-  ram_size = df_ram_size()
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - -
    df

    a disk.frame or the file size in bytes of a CSV file holding the -data

    type

    only = "csv" is supported. It indicates the file type -corresponding to file size `df`

    minchunks

    the minimum number of chunks. Defaults to the number of CPU -cores (without hyper-threading)

    conservatism

    a multiplier to the recommended number of chunks. The +

    + 
    recommend_nchunks(
+  df,
+  type = "csv",
+  minchunks = data.table::getDTthreads(),
+  conservatism = 8,
+  ram_size = df_ram_size()
+)
    +
    + +
    +

    Arguments

    +
    df
    +

    a disk.frame or the file size in bytes of a CSV file holding the +data

    +
    type
    +

    only = "csv" is supported. It indicates the file type +corresponding to file size `df`

    +
    minchunks
    +

    the minimum number of chunks. Defaults to the number of CPU +cores (without hyper-threading)

    +
    conservatism
    +

    a multiplier to the recommended number of chunks. The more chunks the smaller the chunk size and more likely that each chunk can -fit into RAM

    ram_size

    The amount of RAM available which is usually computed. Except on RStudio with R3.6+

    - - -

    Examples

    - 
    # recommend nchunks based on data.frame
-recommend_nchunks(cars)
-
    #> [1] 6
    
-# recommend nchunks based on file size ONLY CSV is implemented at the moment
-recommend_nchunks(1024^3)
-
    #> [1] 6
    +fit into RAM

    +
    ram_size
    +

    The amount of RAM available which is usually computed. Except on RStudio with R3.6+

    +
    + +
    +

    Examples

    + 
    # recommend nchunks based on data.frame
+recommend_nchunks(cars)
+#> [1] 6
+
+# recommend nchunks based on file size ONLY CSV is implemented at the moment
+recommend_nchunks(1024^3)
+#> [1] 6
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/remove_chunk.html b/docs/reference/remove_chunk.html index 6e5db636..974ae319 100644 --- a/docs/reference/remove_chunk.html +++ b/docs/reference/remove_chunk.html @@ -1,67 +1,12 @@ - - - - - - - -Removes a chunk from the disk.frame — remove_chunk • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Removes a chunk from the disk.frame — remove_chunk • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Removes a chunk from the disk.frame

    - Source: R/remove_chunk.r + Source: R/remove_chunk.r
    @@ -165,85 +95,83 @@

    Removes a chunk from the disk.frame

    Removes a chunk from the disk.frame

    - 
    remove_chunk(df, chunk_id, full.names = FALSE)
    - -

    Arguments

    - - - - - - - - - - - - - - -
    df

    a disk.frame

    chunk_id

    the chunk ID of the chunk to remove. If it's a number then return number.fst

    full.names

    TRUE or FALSE. Defaults to FALSE. If true then chunk_id is the full path to the chunk otherwise it's the relative path

    - - -

    Examples

    - 
    # TODO add these to tests
-cars.df = as.disk.frame(cars, nchunks = 4)
-
-# removes 3rd chunk
-remove_chunk(cars.df, 3)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1473bc5984.df"
-#> nchunks: 3
-#> nrow (at source): 37
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    nchunks(cars.df) # 3
-
    #> [1] 3
    
-# removes 4th chunk
-remove_chunk(cars.df, "4.fst")
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1473bc5984.df"
-#> nchunks: 2
-#> nrow (at source): 26
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    nchunks(cars.df) # 3
-
    #> [1] 2
    
-# removes 2nd chunk
-remove_chunk(cars.df, file.path(attr(cars.df, "path", exact=TRUE), "2.fst"), full.names = TRUE)
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b1473bc5984.df"
-#> nchunks: 1
-#> nrow (at source): 13
-#> ncol (at source): 2
-#> nrow (post operations): ???
-#> ncol (post operations): ???
    nchunks(cars.df) # 1
-
    #> [1] 1
    
-# clean up cars.df
-delete(cars.df)
-
    +
    + 
    remove_chunk(df, chunk_id, full.names = FALSE)
    +
    + +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    chunk_id
    +

    the chunk ID of the chunk to remove. If it's a number then return number.fst

    +
    full.names
    +

    TRUE or FALSE. Defaults to FALSE. If true then chunk_id is the full path to the chunk otherwise it's the relative path

    +
    + +
    +

    Examples

    + 
    # TODO add these to tests
+cars.df = as.disk.frame(cars, nchunks = 4)
+
+# removes 3rd chunk
+remove_chunk(cars.df, 3)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5\file56f45e4b14bd.df"
+#> nchunks: 3
+#> nrow (at source): 37
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+nchunks(cars.df) # 3
+#> [1] 3
+
+# removes 4th chunk
+remove_chunk(cars.df, "4.fst")
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5\file56f45e4b14bd.df"
+#> nchunks: 2
+#> nrow (at source): 26
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+nchunks(cars.df) # 3
+#> [1] 2
+
+# removes 2nd chunk
+remove_chunk(cars.df, file.path(attr(cars.df, "path", exact=TRUE), "2.fst"), full.names = TRUE)
+#> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5\file56f45e4b14bd.df"
+#> nchunks: 1
+#> nrow (at source): 13
+#> ncol (at source): 2
+#> nrow (post operations): ???
+#> ncol (post operations): ???
+nchunks(cars.df) # 1
+#> [1] 1
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/sample.html b/docs/reference/sample.html index 83e4f100..719d4244 100644 --- a/docs/reference/sample.html +++ b/docs/reference/sample.html @@ -1,67 +1,12 @@ - - - - - - - -Sample n rows from a disk.frame — sample_frac.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Sample n rows from a disk.frame — sample_frac.disk.frame • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Sample n rows from a disk.frame

    - Source: R/sample_frac.R + Source: R/sample_frac.R
    @@ -165,100 +95,87 @@

    Sample n rows from a disk.frame

    Sample n rows from a disk.frame

    - 
    # S3 method for disk.frame
-sample_frac(tbl, size = 1, replace = FALSE, weight = NULL, .env = NULL, ...)
    +
    + 
    # S3 method for disk.frame
+sample_frac(tbl, size = 1, replace = FALSE, weight = NULL, .env = NULL, ...)
    +
    -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - -
    tbl

    A data.frame.

    size

    <tidy-select> -For sample_n(), the number of rows to select. -For sample_frac(), the fraction of rows to select. -If tbl is grouped, size applies to each group.

    replace

    Sample with or without replacement?

    weight

    <tidy-select> Sampling weights. +

    +

    Arguments

    +
    tbl
    +

    A data.frame.

    +
    size
    +

    <tidy-select> +For sample_n(), the number of rows to select. +For sample_frac(), the fraction of rows to select. +If tbl is grouped, size applies to each group.

    +
    replace
    +

    Sample with or without replacement?

    +
    weight
    +

    <tidy-select> Sampling weights. This must evaluate to a vector of non-negative numbers the same length as -the input. Weights are automatically standardised to sum to 1.

    .env

    DEPRECATED.

    ...

    ignored

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
-
-collect(sample_frac(cars.df, 0.5))
-
    #>     speed dist
-#>  1:    10   34
-#>  2:     7   22
-#>  3:     8   16
-#>  4:     4    2
-#>  5:    11   17
-#>  6:    12   20
-#>  7:    11   28
-#>  8:    13   34
-#>  9:    15   26
-#> 10:    15   54
-#> 11:    15   20
-#> 12:    14   80
-#> 13:    16   40
-#> 14:    18   42
-#> 15:    17   40
-#> 16:    19   36
-#> 17:    19   68
-#> 18:    20   64
-#> 19:    20   52
-#> 20:    20   48
-#> 21:    25   85
-#> 22:    24  120
-#>     speed dist
    
-# clean up cars.df
-delete(cars.df)
-
    +the input. Weights are automatically standardised to sum to 1.

    +
    .env
    +

    DEPRECATED.

    +
    ...
    +

    ignored

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+collect(sample_frac(cars.df, 0.5))
+#>     speed dist
+#>  1:     7   22
+#>  2:     7    4
+#>  3:     9   10
+#>  4:    10   26
+#>  5:    13   34
+#>  6:    11   17
+#>  7:    12   20
+#>  8:    11   28
+#>  9:    14   26
+#> 10:    14   36
+#> 11:    15   26
+#> 12:    14   80
+#> 13:    18   84
+#> 14:    18   56
+#> 15:    18   76
+#> 16:    17   50
+#> 17:    20   56
+#> 18:    19   46
+#> 19:    19   68
+#> 20:    20   32
+#> 21:    25   85
+#> 22:    24   70
+#>     speed dist
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/setup_disk.frame.html b/docs/reference/setup_disk.frame.html index 477f1e08..4b1ee1c7 100644 --- a/docs/reference/setup_disk.frame.html +++ b/docs/reference/setup_disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -Set up disk.frame environment — setup_disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Set up disk.frame environment — setup_disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    Set up disk.frame environment

    - Source: R/setup.r + Source: R/setup.r
    @@ -165,78 +95,71 @@

    Set up disk.frame environment

    Set up disk.frame environment

    - 
    setup_disk.frame(
-  workers = data.table::getDTthreads(),
-  future_backend = future::multisession,
-  ...,
-  gui = FALSE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    workers

    the number of workers (background R processes in the

    future_backend

    which future backend to use for parallelization

    ...

    passed to `future::plan`

    gui

    Whether to use a Graphical User Interface (GUI) for selecting the options. Defaults to FALSE

    - - -

    Examples

    - 
    if (interactive()) {
-  # setup disk.frame to use multiple workers these may use more than two
-  # cores, and is therefore not allowed on CRAN. Hence it's set to run only in
-  # interactive session
-  setup_disk.frame()
-  
-  # use a Shiny GUI to adjust settings
-  # only run in interactive()
-  setup_disk.frame(gui = TRUE)
-}
-
-# set the number workers to 2
-setup_disk.frame(2)
-
    #> The number of workers available for disk.frame is 2
    
-# if you do not wish to use multiple workers you can set it to sequential
-setup_disk.frame(future_backend=future::sequential)
-
    #> Warning: Detected 1 unknown future arguments: 'workers'
    #> The number of workers available for disk.frame is 1
    +
    + 
    setup_disk.frame(
+  workers = data.table::getDTthreads(),
+  future_backend = future::multisession,
+  ...,
+  gui = FALSE
+)
    +
    + +
    +

    Arguments

    +
    workers
    +

    the number of workers (background R processes in the

    +
    future_backend
    +

    which future backend to use for parallelization

    +
    ...
    +

    passed to `future::plan`

    +
    gui
    +

    Whether to use a Graphical User Interface (GUI) for selecting the options. Defaults to FALSE

    +
    + +
    +

    Examples

    + 
    if (interactive()) {
+  # setup disk.frame to use multiple workers these may use more than two
+  # cores, and is therefore not allowed on CRAN. Hence it's set to run only in
+  # interactive session
+  setup_disk.frame()
+  
+  # use a Shiny GUI to adjust settings
+  # only run in interactive()
+  setup_disk.frame(gui = TRUE)
+}
+
+# set the number workers to 2
+setup_disk.frame(2)
+#> The number of workers available for disk.frame is 2
+
+# if you do not wish to use multiple workers you can set it to sequential
+setup_disk.frame(future_backend=future::sequential)
+#> Warning: Detected 1 unknown future arguments: 'workers'
+#> The number of workers available for disk.frame is 1
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/shard.html b/docs/reference/shard.html index 6e4ed76b..eca41f84 100644 --- a/docs/reference/shard.html +++ b/docs/reference/shard.html @@ -1,68 +1,13 @@ - - - - - - - -Shard a data.frame/data.table or disk.frame into chunk and saves it into a disk.frame — shard • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Shard a data.frame/data.table or disk.frame into chunk and saves it into a disk.frame — shard • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Shard a data.frame/data.table or disk.frame into chunk and saves it into a disk.frame

    - Source: R/shard.r + Source: R/shard.r
    @@ -167,95 +97,76 @@

    Shard a data.frame/data.table or disk.frame into chunk and saves it into a d

    `distribute` is an alias for `shard`

    - 
    shard(
-  df,
-  shardby,
-  outdir = tempfile(fileext = ".df"),
-  ...,
-  nchunks = recommend_nchunks(df),
-  overwrite = FALSE,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL
-)
-
-distribute(...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df

    A data.frame/data.table or disk.frame. If disk.frame, then rechunk(df, ...) is run

    shardby

    The column(s) to shard the data by.

    outdir

    The output directory of the disk.frame

    ...

    not used

    nchunks

    The number of chunks

    overwrite

    If TRUE then the chunks are overwritten

    shardby_function

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    sort_splits

    If shardby_function is "sort", the split values for sharding

    desc_vars

    for the "sort" shardby function, the variables to sort descending.

    - +
    + 
    shard(
+  df,
+  shardby,
+  outdir = tempfile(fileext = ".df"),
+  ...,
+  nchunks = recommend_nchunks(df),
+  overwrite = FALSE,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL
+)
+
+distribute(...)
    +
    -

    Examples

    - 
    
-# shard the cars data.frame by speed so that rows with the same speed are in the same chunk
-iris.df = shard(iris, "Species")
-
    #> Hashing...
    
-# clean up cars.df
-delete(iris.df)
-
    +
    +

    Arguments

    +
    df
    +

    A data.frame/data.table or disk.frame. If disk.frame, then rechunk(df, ...) is run

    +
    shardby
    +

    The column(s) to shard the data by.

    +
    outdir
    +

    The output directory of the disk.frame

    +
    ...
    +

    not used

    +
    nchunks
    +

    The number of chunks

    +
    overwrite
    +

    If TRUE then the chunks are overwritten

    +
    shardby_function
    +

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    +
    sort_splits
    +

    If shardby_function is "sort", the split values for sharding

    +
    desc_vars
    +

    for the "sort" shardby function, the variables to sort descending.

    +
    + +
    +

    Examples

    + 
    
+# shard the cars data.frame by speed so that rows with the same speed are in the same chunk
+iris.df = shard(iris, "Species")
+#> Hashing...
+
+# clean up cars.df
+delete(iris.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/shardkey.html b/docs/reference/shardkey.html index a23b17ef..eeca73d7 100644 --- a/docs/reference/shardkey.html +++ b/docs/reference/shardkey.html @@ -1,67 +1,12 @@ - - - - - - - -Returns the shardkey (not implemented yet) — shardkey • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Returns the shardkey (not implemented yet) — shardkey • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Returns the shardkey (not implemented yet)

    - Source: R/shardkey.r + Source: R/shardkey.r
    @@ -165,43 +95,37 @@

    Returns the shardkey (not implemented yet)

    Returns the shardkey (not implemented yet)

    - 
    shardkey(df)
    - -

    Arguments

    - - - - - - -
    df

    a disk.frame

    +
    + 
    shardkey(df)
    +
    +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/shardkey_equal.html b/docs/reference/shardkey_equal.html index 14596490..e51669c5 100644 --- a/docs/reference/shardkey_equal.html +++ b/docs/reference/shardkey_equal.html @@ -1,67 +1,12 @@ - - - - - - - -Compare two disk.frame shardkeys — shardkey_equal • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Compare two disk.frame shardkeys — shardkey_equal • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Compare two disk.frame shardkeys

    - Source: R/shardkey.r + Source: R/shardkey.r
    @@ -165,47 +95,39 @@

    Compare two disk.frame shardkeys

    Compare two disk.frame shardkeys

    - 
    shardkey_equal(sk1, sk2)
    - -

    Arguments

    - - - - - - - - - - -
    sk1

    shardkey1

    sk2

    shardkey2

    +
    + 
    shardkey_equal(sk1, sk2)
    +
    +
    +

    Arguments

    +
    sk1
    +

    shardkey1

    +
    sk2
    +

    shardkey2

    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/show_ceremony.html b/docs/reference/show_ceremony.html index f3db3865..a4e1badc 100644 --- a/docs/reference/show_ceremony.html +++ b/docs/reference/show_ceremony.html @@ -1,67 +1,12 @@ - - - - - - - -Show the code to setup disk.frame — show_ceremony • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Show the code to setup disk.frame — show_ceremony • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Show the code to setup disk.frame

    - Source: R/show_ceremony.R + Source: R/show_ceremony.R
    @@ -165,41 +95,38 @@

    Show the code to setup disk.frame

    Show the code to setup disk.frame

    - 
    show_ceremony()
+    
    
+    
    show_ceremony()
 
-ceremony_text()
+ceremony_text()
 
-show_boilerplate()
-
-insert_ceremony()
    
+show_boilerplate()
 
+insert_ceremony()
    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/srckeep.html b/docs/reference/srckeep.html index 898b76fa..934352ba 100644 --- a/docs/reference/srckeep.html +++ b/docs/reference/srckeep.html @@ -1,67 +1,12 @@ - - - - - - - -Keep only the variables from the input listed in selections — srckeep • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Keep only the variables from the input listed in selections — srckeep • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Keep only the variables from the input listed in selections

    - Source: R/srckeep.disk.frame.r + Source: R/srckeep.disk.frame.r
    @@ -165,117 +95,108 @@

    Keep only the variables from the input listed in selections

    Keep only the variables from the input listed in selections

    - 
    srckeep(diskf, selections, ...)
-
-srckeepchunks(diskf, chunks, ...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - -
    diskf

    a disk.frame

    selections

    The list of variables to keep from the input source

    ...

    not yet used

    chunks

    The chunks to load

    - +
    + 
    srckeep(diskf, selections, ...)
 
-    
    Examples
    
-    
    cars.df = as.disk.frame(cars)
+srckeepchunks(diskf, chunks, ...)
    +
    -# when loading cars's chunks into RAM, load only the column speed -collect(srckeep(cars.df, "speed")) -
    #> speed -#> 1: 4 -#> 2: 4 -#> 3: 7 -#> 4: 7 -#> 5: 8 -#> 6: 9 -#> 7: 10 -#> 8: 10 -#> 9: 10 -#> 10: 11 -#> 11: 11 -#> 12: 12 -#> 13: 12 -#> 14: 12 -#> 15: 12 -#> 16: 13 -#> 17: 13 -#> 18: 13 -#> 19: 13 -#> 20: 14 -#> 21: 14 -#> 22: 14 -#> 23: 14 -#> 24: 15 -#> 25: 15 -#> 26: 15 -#> 27: 16 -#> 28: 16 -#> 29: 17 -#> 30: 17 -#> 31: 17 -#> 32: 18 -#> 33: 18 -#> 34: 18 -#> 35: 18 -#> 36: 19 -#> 37: 19 -#> 38: 19 -#> 39: 20 -#> 40: 20 -#> 41: 20 -#> 42: 20 -#> 43: 20 -#> 44: 22 -#> 45: 23 -#> 46: 24 -#> 47: 24 -#> 48: 24 -#> 49: 24 -#> 50: 25 -#> speed
    -# clean up cars.df -delete(cars.df) -
    +
    +

    Arguments

    +
    diskf
    +

    a disk.frame

    +
    selections
    +

    The list of variables to keep from the input source

    +
    ...
    +

    not yet used

    +
    chunks
    +

    The chunks to load

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# when loading cars's chunks into RAM, load only the column speed
+collect(srckeep(cars.df, "speed"))
+#>     speed
+#>  1:     4
+#>  2:     4
+#>  3:     7
+#>  4:     7
+#>  5:     8
+#>  6:     9
+#>  7:    10
+#>  8:    10
+#>  9:    10
+#> 10:    11
+#> 11:    11
+#> 12:    12
+#> 13:    12
+#> 14:    12
+#> 15:    12
+#> 16:    13
+#> 17:    13
+#> 18:    13
+#> 19:    13
+#> 20:    14
+#> 21:    14
+#> 22:    14
+#> 23:    14
+#> 24:    15
+#> 25:    15
+#> 26:    15
+#> 27:    16
+#> 28:    16
+#> 29:    17
+#> 30:    17
+#> 31:    17
+#> 32:    18
+#> 33:    18
+#> 34:    18
+#> 35:    18
+#> 36:    19
+#> 37:    19
+#> 38:    19
+#> 39:    20
+#> 40:    20
+#> 41:    20
+#> 42:    20
+#> 43:    20
+#> 44:    22
+#> 45:    23
+#> 46:    24
+#> 47:    24
+#> 48:    24
+#> 49:    24
+#> 50:    25
+#>     speed
+
+# clean up cars.df
+delete(cars.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/sub-.disk.frame.html b/docs/reference/sub-.disk.frame.html index 40e28919..b3ecb1b5 100644 --- a/docs/reference/sub-.disk.frame.html +++ b/docs/reference/sub-.disk.frame.html @@ -1,67 +1,12 @@ - - - - - - - -[ interface for disk.frame using fst backend — [.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -[ interface for disk.frame using fst backend — [.disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    [ interface for disk.frame using fst backend

    - Source: R/data.table.r + Source: R/data.table.r
    @@ -165,7 +95,8 @@

    [ interface for disk.frame using fst backend

    [ interface for disk.frame using fst backend

    - 
    # S3 method for disk.frame
+    
    
+    
    # S3 method for disk.frame
 [(
   df,
   ...,
@@ -174,75 +105,59 @@ 
    [ interface for disk.frame using fst backend
    
   use.names = TRUE,
   fill = FALSE,
   idcol = NULL
-)
    
-
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
-      
-      
-    
-    
    df
    a disk.frame
    ...
    same as data.table
    keep
    the columns to srckeep
    rbind
    Whether to rbind the chunks. Defaults to TRUE
    use.names
    Same as in data.table::rbindlist
    fill
    Same as in data.table::rbindlist
    idcol
    Same as in data.table::rbindlist
    
-
-
-    
    Examples
    
-    
    cars.df = as.disk.frame(cars)
-speed_limit = 50
-cars.df[speed < speed_limit ,.N, cut(dist, pretty(dist))]
-
    #> Error in .checkTypos(e, names_x): Object 'speed_limit' not found amongst speed, dist
    
-# clean up
-delete(cars.df)
-
    
+)
    +
    + +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    ...
    +

    same as data.table

    +
    keep
    +

    the columns to srckeep

    +
    rbind
    +

    Whether to rbind the chunks. Defaults to TRUE

    +
    use.names
    +

    Same as in data.table::rbindlist

    +
    fill
    +

    Same as in data.table::rbindlist

    +
    idcol
    +

    Same as in data.table::rbindlist

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+speed_limit = 50
+cars.df[speed < speed_limit ,.N, cut(dist, pretty(dist))]
+#> Error in .checkTypos(e, names_x): Object 'speed_limit' not found amongst speed, dist
+
+# clean up
+delete(cars.df)
+
    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/tbl_vars.disk.frame.html b/docs/reference/tbl_vars.disk.frame.html index 4869fa2f..d61fa7a6 100644 --- a/docs/reference/tbl_vars.disk.frame.html +++ b/docs/reference/tbl_vars.disk.frame.html @@ -1,68 +1,13 @@ - - - - - - - -Column names for RStudio auto-complete — tbl_vars.disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Column names for RStudio auto-complete — tbl_vars.disk.frame • disk.frame - - - - + + -
    -
    - -
    - -
    +

    Column names for RStudio auto-complete

    - Source: R/tbl_vars.r + Source: R/tbl_vars.r
    @@ -167,47 +97,41 @@

    Column names for RStudio auto-complete

    names

    - 
    # S3 method for disk.frame
-tbl_vars(x)
+    
    
+    
    # S3 method for disk.frame
+tbl_vars(x)
 
-# S3 method for disk.frame
-group_vars(x)
    
-
-    
    Arguments
    
-    
-    
-    
-      
-      
-    
-    
    x
    a disk.frame
    
+# S3 method for disk.frame
+group_vars(x)
    +
    +
    +

    Arguments

    +
    x
    +

    a disk.frame

    +
    + -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/write_disk.frame.html b/docs/reference/write_disk.frame.html index 34f8bdf9..5f3f9fdb 100644 --- a/docs/reference/write_disk.frame.html +++ b/docs/reference/write_disk.frame.html @@ -1,68 +1,13 @@ - - - - - - - -Write disk.frame to disk — write_disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -Write disk.frame to disk — write_disk.frame • disk.frame + + - - - - -
    -
    - -
    - -
    +

    Write disk.frame to disk

    - Source: R/write_disk.frame.r + Source: R/write_disk.frame.r
    @@ -167,110 +97,89 @@

    Write disk.frame to disk

    then using the as.disk.frame function is recommended for most cases

    - 
    write_disk.frame(
-  df,
-  outdir = tempfile(fileext = ".df"),
-  nchunks = ifelse("disk.frame" %in% class(df), nchunks.disk.frame(df),
-    recommend_nchunks(df)),
-  overwrite = FALSE,
-  shardby = NULL,
-  compress = 50,
-  shardby_function = "hash",
-  sort_splits = NULL,
-  desc_vars = NULL,
-  ...
-)
-
-output_disk.frame(...)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    df

    a disk.frame

    outdir

    output directory for the disk.frame

    nchunks

    number of chunks

    overwrite

    overwrite output directory

    shardby

    the columns to shard by

    compress

    compression ratio for fst files

    shardby_function

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    sort_splits

    for the "sort" shardby function, a dataframe with the split values.

    desc_vars

    for the "sort" shardby function, the variables to sort descending.

    ...

    passed to cmap.disk.frame

    - - -

    Examples

    - 
    cars.df = as.disk.frame(cars)
+    
    
+    
    write_disk.frame(
+  df,
+  outdir = tempfile(fileext = ".df"),
+  nchunks = ifelse("disk.frame" %in% class(df), nchunks.disk.frame(df),
+    recommend_nchunks(df)),
+  overwrite = FALSE,
+  shardby = NULL,
+  compress = 50,
+  shardby_function = "hash",
+  sort_splits = NULL,
+  desc_vars = NULL,
+  ...
+)
+
+output_disk.frame(...)
    
+    
    
 
-# write out a lazy disk.frame to disk
-cars2.df = write_disk.frame(cmap(cars.df, ~.x[1,]), overwrite = TRUE)
-collect(cars2.df)
-
    #>    speed dist
-#> 1:     4    2
-#> 2:    11   17
-#> 3:    13   46
-#> 4:    16   40
-#> 5:    19   46
-#> 6:    24   70
    
-# clean up cars.df
-delete(cars.df)
-delete(cars2.df)
-
    +
    +

    Arguments

    +
    df
    +

    a disk.frame

    +
    outdir
    +

    output directory for the disk.frame

    +
    nchunks
    +

    number of chunks

    +
    overwrite
    +

    overwrite output directory

    +
    shardby
    +

    the columns to shard by

    +
    compress
    +

    compression ratio for fst files

    +
    shardby_function
    +

    splitting of chunks: "hash" for hash function or "sort" for semi-sorted chunks

    +
    sort_splits
    +

    for the "sort" shardby function, a dataframe with the split values.

    +
    desc_vars
    +

    for the "sort" shardby function, the variables to sort descending.

    +
    ...
    +

    passed to cmap.disk.frame

    +
    + +
    +

    Examples

    + 
    cars.df = as.disk.frame(cars)
+
+# write out a lazy disk.frame to disk
+cars2.df = write_disk.frame(cmap(cars.df, ~.x[1,]), overwrite = TRUE)
+collect(cars2.df)
+#>    speed dist
+#> 1:     4    2
+#> 2:    11   17
+#> 3:    13   46
+#> 4:    16   40
+#> 5:    19   46
+#> 6:    24   70
+
+# clean up cars.df
+delete(cars.df)
+delete(cars2.df)
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/reference/zip_to_disk.frame.html b/docs/reference/zip_to_disk.frame.html index a6f1a16e..9188f8ac 100644 --- a/docs/reference/zip_to_disk.frame.html +++ b/docs/reference/zip_to_disk.frame.html @@ -1,70 +1,15 @@ - - - - - - - -`zip_to_disk.frame` is used to read and convert every CSV file within the zip -file to disk.frame format — zip_to_disk.frame • disk.frame - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -`zip_to_disk.frame` is used to read and convert every CSV file within the zip +file to disk.frame format — zip_to_disk.frame • disk.frame - - + + - - -
    -
    - -
    - -
    +

    `zip_to_disk.frame` is used to read and convert every CSV file within the zip file to disk.frame format

    - Source: R/zip_to_disk.frame.r + Source: R/zip_to_disk.frame.r
    @@ -170,95 +100,84 @@

    `zip_to_disk.frame` is used to read and convert every CSV file within the zi file to disk.frame format

    - 
    zip_to_disk.frame(
-  zipfile,
-  outdir,
-  ...,
-  validation.check = FALSE,
-  overwrite = TRUE
-)
    - -

    Arguments

    - - - - - - - - - - - - - - - - - - - - - - -
    zipfile

    The zipfile

    outdir

    The output directory for disk.frame

    ...

    passed to fread

    validation.check

    should the function perform a check at the end to check for validity of output. It can detect issues with conversion

    overwrite

    overwrite output directory

    - -

    Value

    +
    + 
    zip_to_disk.frame(
+  zipfile,
+  outdir,
+  ...,
+  validation.check = FALSE,
+  overwrite = TRUE
+)
    +
    +
    +

    Arguments

    +
    zipfile
    +

    The zipfile

    +
    outdir
    +

    The output directory for disk.frame

    +
    ...
    +

    passed to fread

    +
    validation.check
    +

    should the function perform a check at the end to check for validity of output. It can detect issues with conversion

    +
    overwrite
    +

    overwrite output directory

    +
    +
    +

    Value

    a list of disk.frame

    -

    See also

    - -

    Other ingesting data: -csv_to_disk.frame()

    - -

    Examples

    - 
    # create a zip file containing a csv
-csvfile = tempfile(fileext = ".csv")
-write.csv(cars, csvfile)
-zipfile = tempfile(fileext = ".zip")
-zip(zipfile, csvfile)
-
-# read every file and convert it to a disk.frame
-zip.df = zip_to_disk.frame(zipfile, tempfile(fileext = ".df"))
+    
    
+    
    
+    
    See also
    
+    
    Other ingesting data: 
+csv_to_disk.frame()
    
+    
    
 
-# there is only one csv file so it return a list of one disk.frame
-zip.df[[1]]
-
    #> path: "C:\Users\RTX2080\AppData\Local\Temp\Rtmp2PGAkP\file4b14559e63cf.df/Users/RTX2080/AppData/Local/Temp/Rtmp2PGAkP/file4b1437ba44b8.csv" -#> nchunks: 6 -#> nrow (at source): 50 -#> ncol (at source): 3 -#> nrow (post operations): ??? -#> ncol (post operations): ???
    -# clean up -unlink(csvfile) -unlink(zipfile) -delete(zip.df[[1]]) -
    +
    +

    Examples

    + 
    # create a zip file containing a csv
+csvfile = tempfile(fileext = ".csv")
+write.csv(cars, csvfile)
+zipfile = tempfile(fileext = ".zip")
+zip(zipfile, csvfile)
+#> Warning: '"zip"' not found
+
+# read every file and convert it to a disk.frame
+zip.df = zip_to_disk.frame(zipfile, tempfile(fileext = ".df"))
+#> Error in unzip(zipfile, list = TRUE): zip file 'C:\Users\RTX2080\AppData\Local\Temp\Rtmp2rQjw5\file56f44b886b42.zip' cannot be opened
+
+# there is only one csv file so it return a list of one disk.frame
+zip.df[[1]]
+#> Error in eval(expr, envir, enclos): object 'zip.df' not found
+
+# clean up
+unlink(csvfile)
+unlink(zipfile)
+delete(zip.df[[1]])
+#> Error in "disk.frame" %in% class(df): object 'zip.df' not found
+
    +
    +
    -
    -
    -

    Developed by Dai ZJ.

    +
    +

    Developed by Dai ZJ.

    -

    Site built with pkgdown 1.6.1.

    +

    Site built with pkgdown 2.0.2.

    -
    -
    +
    - - + + diff --git a/docs/sitemap.xml b/docs/sitemap.xml new file mode 100644 index 00000000..4aa93c2a --- /dev/null +++ b/docs/sitemap.xml @@ -0,0 +1,294 @@ + + + + /02-intro-disk-frame.html + + + /404.html + + + /articles/01-intro-disk-frame.html + + + /articles/01-intro.html + + + /articles/02-common-questions.html + + + /articles/02-intro-disk-frame.html + + + /articles/03-concepts.html + + + /articles/03_concepts.html + + + /articles/04-ingesting-data.html + + + /articles/04_ingesting-data.html + + + /articles/05-data-table-syntax.html + + + /articles/06-vs-dask-juliadb.html + + + /articles/07-glm.html + + + /articles/08-more-epic.html + + + /articles/09-convenience-features.html + + + /articles/10-group-by.html + + + /articles/11-custom-group-by.html + + + /articles/88-trouble-shooting.html + + + /articles/common-questions.html + + + /articles/concepts.html + + + /articles/convenience-features.html + + + /articles/custom-group-by.html + + + /articles/data-table-syntax.html + + + /articles/glm.html + + + /articles/group-by.html + + + /articles/index.html + + + /articles/ingesting-data.html + + + /articles/intro-disk-frame.html + + + /articles/intro.html + + + /articles/more-epic.html + + + /articles/vs-dask-juliadb-2.html + + + /articles/vs-dask-juliadb.html + + + /authors.html + + + /index.html + + + /LICENSE-text.html + + + /news/index.html + + + /reference/add_chunk.html + + + /reference/as.data.frame.disk.frame.html + + + /reference/as.data.table.disk.frame.html + + + /reference/as.disk.frame.html + + + /reference/bloomfilter.html + + + /reference/chunk_group_by.html + + + /reference/cmap.html + + + /reference/cmap2.html + + + /reference/collect.html + + + /reference/colnames.html + + + /reference/compute.disk.frame.html + + + /reference/create_chunk_mapper.html + + + /reference/create_dplyr_mapper.html + + + /reference/csv_to_disk.frame.html + + + /reference/delete.html + + + /reference/dfglm.html + + + /reference/df_ram_size.html + + + /reference/disk.frame.html + + + /reference/dplyr_verbs.html + + + /reference/evalparseglue.html + + + /reference/foverlaps.disk.frame.html + + + /reference/gen_datatable_synthetic.html + + + /reference/get_chunk.html + + + /reference/get_chunk_ids.html + + + /reference/groups.disk.frame.html + + + /reference/group_by.html + + + /reference/hard_arrange.html + + + /reference/hard_group_by.html + + + /reference/head_tail.html + + + /reference/index.html + + + /reference/is_disk.frame.html + + + /reference/join.html + + + /reference/make_glm_streaming_fn.html + + + /reference/map.html + + + /reference/map2.html + + + /reference/mean.chunk_agg.disk.frame.html + + + /reference/mean.collected_agg.disk.frame.html + + + /reference/merge.disk.frame.html + + + /reference/move_to.html + + + /reference/nchunks.html + + + /reference/ncol_nrow.html + + + /reference/one-stage-group-by-verbs.html + + + /reference/overwrite_check.html + + + /reference/print.disk.frame.html + + + /reference/pull.disk.frame.html + + + /reference/rbindlist.disk.frame.html + + + /reference/rechunk.html + + + /reference/recommend_nchunks.html + + + /reference/remove_chunk.html + + + /reference/sample.html + + + /reference/setup_disk.frame.html + + + /reference/shard.html + + + /reference/shardkey.html + + + /reference/shardkey_equal.html + + + /reference/show_ceremony.html + + + /reference/srckeep.html + + + /reference/sub-.disk.frame.html + + + /reference/summarise.grouped_disk.frame.html + + + /reference/tbl_vars.disk.frame.html + + + /reference/write_disk.frame.html + + + /reference/zip_to_disk.frame.html + +