Merge pull request #6 from CDCgov/V1.6-2024-03-11-run

V1.6 2024 03 11 run

.gitignore

@@ -18,14 +18,8 @@ secrets.yaml
 *.pdf
 *.docx
 
-# Don't track compiled model executable
-
-cfaforecastrenewalww/inst/stan/renewal_ww_hosp
-cfaforecastrenewalww/inst/stan/renewal_ww_hosp_hierarchical_w_dispersion
-cfaforecastrenewalww/inst/stan/renewal_ww_hosp_site_level_phi
-cfaforecastrenewalww/inst/stan/renewal_ww_hosp_site_level_phi_varying_C
-cfaforecastrenewalww/inst/stan/renewal_ww_hosp_hierarchical
-cfaforecastrenewalww/inst/stan/renewal_ww_hosp_site_level_inf_dynamics
+# Don't track compiled model executables
+bin
 
 # Don't track data by default
 repo_data/
@@ -37,10 +31,14 @@ output/*
 !input/locations.csv
 !input/params.toml
 
-# Do track the output/forecasts folder,
+# Do track the output/forecasts directory
 !output/forecasts
 !output/forecasts/*
 
+# Do track the output/decision_archive directory
+!output/decision_archive
+!output/decision_archive/*
+
 # compiled test executables
 test/test_expgamma_lpdf
 

README.md

@@ -203,6 +203,8 @@ Alternatively, in an interactive R session with your R working directory set to
 targets::tar_make()
 ```
 
+The first time you run the pipeline after installing `cfaforecastrenewalww`, `cmdstan` will compile the model source `.stan` files to executible binaries, which by default are stored in a subdirectory `bin/` of the top-level project directory. Subsequent runs should use those precompiled executibles, without need for recompilation. To force recompilation, delete the binaries stored in the `bin/` directory or reinstall the `cfaforecastrenewalww` R package.
+
 # Contributing to this project
 
 ## Git workflow

_targets.R

@@ -220,8 +220,10 @@ list(
     priority = 1
   ),
   tar_target(
-    name = model_file_id,
-    command = ww_model(model_file_path_id),
+    name = model_object_id,
+    command = compile_model(model_file_path_id,
+      target_dir = "bin"
+    ),
     deployment = "main",
     priority = 1
   ),
@@ -256,14 +258,16 @@ list(
     name = df_of_filepaths_id,
     command = do.call(
       fit_site_level_model,
-      c(list(train_data = grouped_data_id),
-        list(params = params_id),
-        model_file = model_file_id,
+      c(
+        list(
+          train_data = grouped_data_id,
+          params = params_id,
+          model_file = model_object_id
+        ),
         config_vars_id
       )
     ),
     pattern = map(grouped_data_id),
-    iteration = "group",
     deployment = "worker",
     priority = 1
   ),
@@ -402,8 +406,10 @@ list(
     deployment = "main"
   ),
   tar_target(
-    name = model_file_ho,
-    command = ww_model(model_file_path_ho),
+    name = model_object_ho,
+    command = compile_model(model_file_path_ho,
+      target_dir = "bin"
+    ),
     deployment = "main"
   ),
   tar_target(
@@ -431,13 +437,18 @@ list(
   # generated quantities and the parameters
   tar_target(
     name = df_of_filepaths_ho,
-    command = do.call(fit_aggregated_model, c(list(train_data = grouped_data),
-      list(params = params_ho),
-      model_file = model_file_ho,
-      config_vars_ho
-    )),
+    command = do.call(
+      fit_aggregated_model,
+      c(
+        list(
+          train_data = grouped_data,
+          params = params_ho,
+          model_file = model_object_ho
+        ),
+        config_vars_ho
+      )
+    ),
     pattern = map(grouped_data),
-    iteration = "group",
     deployment = "worker"
   ),
   tar_target(
@@ -526,8 +537,10 @@ list(
     deployment = "main"
   ),
   tar_target(
-    name = model_file_sa,
-    command = ww_model(model_file_path_sa),
+    name = model_object_sa,
+    command = compile_model(model_file_path_sa,
+      target_dir = "bin"
+    ),
     deployment = "main"
   ),
   tar_target(
@@ -562,13 +575,18 @@ list(
   # generated quantities and the parameters
   tar_target(
     name = df_of_filepaths_sa,
-    command = do.call(fit_aggregated_model, c(list(train_data = grouped_data_sa),
-      list(params = params_sa),
-      model_file = model_file_sa,
-      config_vars_sa
-    )),
+    command = do.call(
+      fit_aggregated_model,
+      c(
+        list(
+          train_data = grouped_data_sa,
+          params = params_sa,
+          model_file = model_object_sa
+        ),
+        config_vars_sa
+      )
+    ),
     pattern = map(grouped_data_sa),
-    iteration = "group",
     deployment = "worker"
   ),
   tar_target(

cfaforecastrenewalww/DESCRIPTION

@@ -57,12 +57,13 @@ Imports:
     posterior,
     scoringutils,
     tidyr,
-    targets,
+    targets (>= 1.5.1),
     tarchetypes,
     gridExtra,
     rlang,
     RcppTOML,
-    RcppEigen
+    RcppEigen,
+    crew (>= 0.9.0)
 VignetteBuilder:
     knitr
 Suggests:

cfaforecastrenewalww/NAMESPACE

@@ -5,6 +5,7 @@ export(aggregate_ww)
 export(boxplot_whiskers)
 export(categorical_entropy)
 export(cleanup_secrets)
+export(compile_model)
 export(convert_to_logmean)
 export(convert_to_logsd)
 export(create_dir)
@@ -109,7 +110,6 @@ export(to_simplex)
 export(trajectories_to_quantiles)
 export(validate_paramlist)
 export(varies)
-export(ww_model)
 export(zoom_boxplot_y)
 import(boot)
 import(cmdstanr)

cfaforecastrenewalww/R/compile_model.R

@@ -0,0 +1,91 @@
+#' This code was adapted from code written
+#' (under an MIT license) as part of the `epinowcast`
+#' package (https://github.com/epinowcast/epinowcast)
+
+#' Compile a stan model while pointing at the package default
+#' include directory (`stan`) for #include statements
+#'
+#' compile_model
+#' @description
+#' This function reads in and optionally compiles a Stan model.
+#' It is written to search the installed package `stan` directory
+#' for additional source files to include. Within each stan file,
+#' use #include {path to your file with the `stan` directory}.stan
+#'
+#' @param model_filepath path to .stan file defining the model
+#' @param include_paths path(s) to directories to search for files
+#' specified in #include statements. Passed to [cmdstanr::cmdstan_model()].
+#' Defaults to the `stan` subdirectory of the installed
+#' `cfaforecastrenewalww` package.
+#' @param threads Number of threads to use in model compilation,
+#' as an integer. Passed to [cmdstanr::cmdstan_model()].
+#' Default `FALSE` (use single-threaded compilation).
+#' @param target_dir Directory in which to save the compiled
+#' stan model binary. Passed as the `dir` keyword argument to
+#' [cmdstanr::cmdstan_model()]. Defaults to a temporary directory
+#' for the R sessions (the output of [tempdir()]).
+#' @param stanc_options Options for the stan compiler passed to
+#' [cmdstanr::cmdstan_model()], as a list. See that function's
+#' documentation for more details. Default `list()` (use default
+#' options).
+#' @param cpp_options Options for the C++ compiler passed to
+#' [cmdstanr::cmdstan_model()], as a list. See that function's
+#' documentation for more details. Default `list()` (use default
+#' options).
+#' @param verbose Write detailed output to the terminal while
+#' executing the function? Boolean, default `TRUE`.
+#' @param ... Additional keyword arguments passed to
+#' [cmdstanr::cmdstan_model()].
+#'
+#' @return The resulting `cmdstanr` model object, as the output
+#' of [cmdstanr::cmdstan_model()].
+#' @export
+compile_model <- function(model_filepath,
+                          include_paths = system.file(
+                            "stan",
+                            package = "cfaforecastrenewalww"
+                          ),
+                          threads = FALSE,
+                          target_dir = tempdir(),
+                          stanc_options = list(),
+                          cpp_options = list(),
+                          verbose = TRUE,
+                          ...) {
+  if (verbose) {
+    cli::cli_inform(
+      glue::glue(paste0(
+        "Using model source file: ",
+        "{model_filepath}"
+      ))
+    )
+    cli::cli_inform(
+      sprintf(
+        "Using include paths: %s",
+        toString(include_paths)
+      )
+    )
+  }
+
+  create_dir(target_dir)
+
+  model <- cmdstanr::cmdstan_model(
+    model_filepath,
+    include_paths = include_paths,
+    compile = TRUE,
+    stanc_options = stanc_options,
+    cpp_options = cpp_options,
+    threads = threads,
+    dir = target_dir,
+    ...
+  )
+
+  if (verbose) {
+    cli::cli_inform(paste0(
+      "Model compiled successfully; ",
+      "model executable binary located at: ",
+      "{model$exe_file()}"
+    ))
+  }
+
+  return(model)
+}

cfaforecastrenewalww/R/fit_model.R

@@ -7,61 +7,6 @@
 #' (under an MIT license) as part of the `epinowcast`
 #' package (https://github.com/epinowcast/epinowcast)
 
-# Compile a stan model and include stan function files
-#' ww_model
-#' @description
-#' This function compiles the stan model, and is written to include the 'stan'
-#' folder. Within each stan file, to include the functions, use #include
-#' functions/{your_function_file}.stan
-#'
-#'
-#' @param model
-#' @param include
-#' @param compile
-#' @param threads
-#' @param target_dir
-#' @param stanc_options
-#' @param cpp_options
-#' @param verbose
-#' @param ...
-#'
-#' @return
-#' @export
-#'
-#' @examples
-ww_model <- function(model,
-                     include = system.file("stan",
-                       package = "cfaforecastrenewalww"
-                     ),
-                     compile = TRUE, threads = FALSE,
-                     target_dir = tempdir(), stanc_options = list(),
-                     cpp_options = list(), verbose = TRUE, ...) {
-  if (verbose) {
-    message(glue::glue("Using model {model}."))
-    message(sprintf("include is %s.", toString(include)))
-  }
-
-  if (compile) {
-    monitor <- suppressMessages
-    if (verbose) {
-      monitor <- function(x) {
-        return(x)
-      }
-    }
-    cpp_options$stan_threads <- threads
-    model <- monitor(cmdstanr::cmdstan_model(
-      model,
-      include_paths = include,
-      stanc_options = stanc_options,
-      cpp_options = cpp_options,
-      ...
-    ))
-  }
-  return(model)
-}
-
-
-
 #' Get dataframe of filepaths
 #'
 #' @param output_dir

cfaforecastrenewalww/R/get_data.R

@@ -181,8 +181,8 @@ get_stan_data <- function(train_data,
     p_hosp_mean = p_hosp_mean,
     p_hosp_sd_logit = p_hosp_sd_logit,
     p_hosp_w_sd_sd = p_hosp_w_sd_sd,
-    infection_feedback_prior_mean = infection_feedback_prior_mean,
-    infection_feedback_prior_sd = infection_feedback_prior_sd
+    inf_feedback_prior_logmean = infection_feedback_prior_logmean,
+    inf_feedback_prior_logsd = infection_feedback_prior_logsd
   )
 
   return(data_renewal)
@@ -475,8 +475,8 @@ get_stan_data_site_level_model <- function(train_data,
     p_hosp_sd_logit = p_hosp_sd_logit,
     p_hosp_w_sd_sd = p_hosp_w_sd_sd,
     ww_site_mod_sd_sd = ww_site_mod_sd_sd,
-    infection_feedback_prior_mean = infection_feedback_prior_mean,
-    infection_feedback_prior_sd = infection_feedback_prior_sd
+    inf_feedback_prior_logmean = infection_feedback_prior_logmean,
+    inf_feedback_prior_logsd = infection_feedback_prior_logsd
   )
 
   if (model_type == "site-level time-varying concentration") {
@@ -551,10 +551,7 @@ state_agg_inits <- function(train_data, params, stan_data) {
     dur_shed = rnorm(1, duration_shedding_mean, 0.1 * duration_shedding_sd),
     log10_g = rnorm(1, log10_g_prior_mean, 0.5),
     hosp_wday_effect = to_simplex(rnorm(7, 1 / 7, 0.01)),
-    infection_feedback = abs(rnorm(
-      1, infection_feedback_prior_mean,
-      0.1 * infection_feedback_prior_sd
-    ))
+    infection_feedback = abs(rnorm(1, 500, 20))
   )
   return(init_list)
 }
@@ -755,10 +752,7 @@ site_level_inf_inits <- function(train_data, params, stan_data) {
     ww_site_mod_raw = abs(rnorm(n_ww_lab_sites, 0, 0.05)),
     ww_site_mod_sd = abs(rnorm(1, 0, 0.05)),
     hosp_wday_effect = to_simplex(rnorm(7, 1 / 7, 0.01)),
-    infection_feedback = abs(rnorm(
-      1, infection_feedback_prior_mean,
-      0.5 * infection_feedback_prior_sd
-    ))
+    infection_feedback = abs(rnorm(1, 500, 20))
   )