bigKRLS is an R algorithm for Kernel-Regularized Least Squares that uses big data packages for size and C++ for speed. This architecture takes a bit of work to set up because R isn't necessarily properly set to connect C++ to C and FORTRAN (i.e., R throws lengthy error messages about lquadmath, lqfortran, clang++, and/or g++) and older compilers aren't necessarily compatabible with parallel processing paradigms that are increasingly popular in the R community, like OMP. Once everything is connected under the hood you'll be able to install R packages like rstan and biglasso too.
bigKRLS has been run on multiple platforms including Mac OS X Yosemite 10.10.5 and Sierra 10.12.6, Linux Ubuntu 14.04, and Windows 7 and Windows 8.
bigKRLS is designed to run on R version 3.3.0 ("Supposedly Educational" released 2016-05-03) or newer. Older, even fairly recent, versions of R will not work with bigmemory.
-- Install the newest R at https://cran.r-project.org
https://cran.r-project.org/bin/windows/Rtools/
RTools best practices
Mac users will need to be sure their compilers are up-to-date. Without fairly current compilers (i.e., compilers that do not come standard with OS X), it will not be possible to install RcppArmadillo.
For the g++ family, version 4.6.* or newer is required. For g++ and related software, see The Coatless Professor's OpenMP in R and OSX. Everything up to the clang4 instructions on that page are recommended.
Mac users will need clang4 (the recent version is required for OMP multicore processing used by other R libraries such as biglasso). Though clang4 can be installed with bash commands, the installer developed by the Coatless Professor (@coatless) is highly recommended since it automatically takes care of the configuration files and paths R requires. For detail, see https://github.com/coatless/r-macos-clang.
If troubles persist, we found the following pages particularly helpful: A, B, and section 2.16 of: C.
To use RStudio, Windows users must use RStudio 1.1.129 or newer and Unix-type users (including Mac) must use 1.0.136 or newer. As of 2017-10-09, the current stable build works for both:
https://www.rstudio.com/products/rstudio/download/
bigKRLS has several dependencies, some of which require recent version of their dependencies. To smooth installation, we recommend installing these packages first.
install.packages(c("Rcpp", "RcppArmadillo", "bigmemory", "biganalytics", "snow", "shiny", "httpuv", "scales", "lazyeval", "tibble"))
You should now be able to install via CRAN.
install.packages("bigKRLS")
library(bigKRLS)
vignette("bigKRLS_basics")
You should now be also able to install bigKRLS via GitHub.
install.packages("devtools")
library(devtools)
Windows users should first run these extra lines:
find_rtools()
find_rtools(T)
Finally, install the most current version with standard devtools syntax:
install_github('rdrr1990/bigKRLS')
library(bigKRLS)
vignette("bigKRLS_basics")
You should be good to go!
Despite improvements, the algorithm is still incredibly memory intensive. We recommend proceeding cautiously and bearing in mind that memory usage is a quadratic function of the number of observations, N (roughly 5N2). Users should estimate models N = 1,000, 2,500, and 5,000 to see how their system performs before considering larger models. See https://sites.google.com/site/petemohanty/software for detail.
Code released under GPL (>= 2).