Merge branch 'tweaks'

- Update version number and change file

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: PLNmodels
 Title: Poisson Lognormal Models
-Version: 1.0.4-0300
+Version: 1.0.5-0000
 Authors@R: c(
   person("Julien", "Chiquet", role = c("aut", "cre"), email = "[email protected]",
       comment = c(ORCID = "0000-0002-3629-3429")),

NEWS.md

@@ -3,6 +3,11 @@
 * Update documentation of PLN*_param() functions to include torch optimization parameters
 * Add (somehow) explicit error message when torch convergence fails
 * Change initialization in `variance_jackknife()` and `variance_bootstrap()` to prevent estimation recycling, results from those functions are now comparable to doing jackknife / bootstrap "by hand". 
+* Merge PR #110 from Cole Trapnell to add:
+  - bootstrap estimation of the variance of model parameter
+  - improved interface for model initialization / optimisation parameters, which
+    are now passed on to jackknife / bootstrap post-treatments
+  - better support of GPU when using torch backend
 
 # PLNmodels 1.0.4 (2023-08-24)
 

R/PLN.R

@@ -45,7 +45,7 @@ PLN <- function(formula, data, subset, weights, control = PLN_param()) {
 
   ## post-treatment
   if (control$trace > 0) cat("\n Post-treatments...")
-  myPLN$postTreatment(args$Y, args$X, args$O, args$w, control$config_post)
+  myPLN$postTreatment(args$Y, args$X, args$O, args$w, control$config_post, control$config_optim)
 
   if (control$trace > 0) cat("\n DONE!\n")
   myPLN
@@ -90,7 +90,7 @@ PLN <- function(formula, data, subset, weights, control = PLN_param()) {
 #' * "etas" pair of multiplicative increase and decrease factors. Default is (0.5, 1.2). Only used in RPROP
 #' * "centered" if TRUE, compute the centered RMSProp where the gradient is normalized by an estimation of its variance weight_decay (L2 penalty). Default to FALSE. Only used in RMSPROP
 #'
-#' The list of parameters `config_post` controls the post-treatment processing (for PLN and PLNLDA), with the following entries:
+#' The list of parameters `config_post` controls the post-treatment processing (for most `PLN*()` functions), with the following entries (defaults may vary depending on the specific function, check `config_post_default_*` for defaults values):
 #' * jackknife boolean indicating whether jackknife should be performed to evaluate bias and variance of the model parameters. Default is FALSE.
 #' * bootstrap integer indicating the number of bootstrap resamples generated to evaluate the variance of the model parameters. Default is 0 (inactivated).
 #' * variational_var boolean indicating whether variational Fisher information matrix should be computed to estimate the variance of the model parameters (highly underestimated). Default is FALSE.

R/PLNLDA.R

@@ -64,7 +64,7 @@ PLNLDA <- function(formula, data, subset, weights, grouping, control = PLN_param
   myLDA$optimize(grouping, args$Y, args$X, args$O, args$w, control$config_optim)
 
   ## Post-treatment: prepare LDA visualization
-  myLDA$postTreatment(grouping, args$Y, args$X, args$O, control$config_post)
+  myLDA$postTreatment(grouping, args$Y, args$X, args$O, control$config_post, control$config_optim)
 
   if (control$trace > 0) cat("\n DONE!\n")
   myLDA

R/PLNLDAfit-class.R

@@ -84,13 +84,14 @@ PLNLDAfit <- R6Class(
     ## %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
     ## Post treatment --------------------
     #' @description  Update R2, fisher and std_err fields and visualization
-    #' @param config list controlling the post-treatment
-    postTreatment = function(grouping, responses, covariates, offsets, config) {
+    #' @param config_post a list for controlling the post-treatments (optional bootstrap, jackknife, R2, etc.).
+    #' @param config_optim list controlling the optimization parameters
+    postTreatment = function(grouping, responses, covariates, offsets, config_post, config_optim) {
       covariates <- cbind(covariates, model.matrix( ~ grouping + 0))
-      super$postTreatment(responses, covariates, offsets, config = config)
+      super$postTreatment(responses, covariates, offsets, config_post = config_post, config_optim = config_optim)
       rownames(private$C) <- colnames(private$C) <- colnames(responses)
       colnames(private$S) <- 1:self$q
-      if (config$trace > 1) cat("\n\tCompute LD scores for visualization...")
+      if (config_post$trace > 1) cat("\n\tCompute LD scores for visualization...")
       self$setVisualization()
     },
 

R/PLNPCA.R

@@ -51,8 +51,7 @@ PLNPCA <- function(formula, data, subset, weights, ranks = 1:5, control = PLNPCA
 
   ## Post-treatments: pseudo-R2, rearrange criteria and prepare PCA visualization
   if (control$trace > 0) cat("\n Post-treatments")
-  config_post <- config_post_default_PLNPCA; config_post$trace <- control$trace
-  myPCA$postTreatment(config_post)
+  myPCA$postTreatment(control$config_post, control$config_optim)
 
   if (control$trace > 0) cat("\n DONE!\n")
   myPCA
@@ -65,6 +64,7 @@ PLNPCA <- function(formula, data, subset, weights, ranks = 1:5, control = PLNPCA
 #' @param backend optimization back used, either "nlopt" or "torch". Default is "nlopt"
 #' @param trace a integer for verbosity.
 #' @param config_optim a list for controlling the optimizer (either "nlopt" or "torch" backend). See details
+#' @param config_post a list for controlling the post-treatments (optional bootstrap, jackknife, R2, etc.). See details
 #' @param inception Set up the parameters initialization: by default, the model is initialized with a multivariate linear model applied on
 #'    log-transformed data, and with the same formula as the one provided by the user. However, the user can provide a PLNfit (typically obtained from a previous fit),
 #'    which sometimes speeds up the inference.
@@ -77,11 +77,17 @@ PLNPCA_param <- function(
     backend       = "nlopt",
     trace         = 1      ,
     config_optim  = list() ,
+    config_post   = list() ,
     inception     = NULL     # pretrained PLNfit used as initialization
 ) {
 
   if (!is.null(inception)) stopifnot(isPLNfit(inception))
 
+  ## post-treatment config
+  config_pst <- config_post_default_PLNPCA
+  config_pst[names(config_post)] <- config_post
+  config_pst$trace <- trace
+
   ## optimization config
   backend <- match.arg(backend)
   stopifnot(backend %in% c("nlopt", "torch"))
@@ -100,5 +106,6 @@ PLNPCA_param <- function(
     backend       = backend   ,
     trace         = trace     ,
     config_optim  = config_opt,
+    config_post   = config_pst,
     inception     = inception   ), class = "PLNmodels_param")
 }

R/PLNPCAfit-class.R

@@ -163,14 +163,16 @@ PLNPCAfit <- R6Class(
       },
 
       #' @description Update R2, fisher, std_err fields and set up visualization
-      #' @details The list of parameters `config` controls the post-treatment processing, with the following entries:
+      #' @param config_optim a list for controlling the optimizer (either "nlopt" or "torch" backend). See details
+      #' @param config_post a list for controlling the post-treatments (optional bootstrap, jackknife, R2, etc.). See details
+      #' @details The list of parameters `config_post` controls the post-treatment processing, with the following entries:
       #' * jackknife boolean indicating whether jackknife should be performed to evaluate bias and variance of the model parameters. Default is FALSE.
       #' * bootstrap integer indicating the number of bootstrap resamples generated to evaluate the variance of the model parameters. Default is 0 (inactivated).
       #' * variational_var boolean indicating whether variational Fisher information matrix should be computed to estimate the variance of the model parameters (highly underestimated). Default is FALSE.
       #' * rsquared boolean indicating whether approximation of R2 based on deviance should be computed. Default is TRUE
       #' * trace integer for verbosity. should be > 1 to see output in post-treatments
-      postTreatment = function(responses, covariates, offsets, weights, config, nullModel) {
-        super$postTreatment(responses, covariates, offsets, weights, config, nullModel)
+      postTreatment = function(responses, covariates, offsets, weights, config_post, config_optim, nullModel) {
+        super$postTreatment(responses, covariates, offsets, weights, config_post, config_optim, nullModel)
         colnames(private$C) <- colnames(private$M) <- 1:self$q
         rownames(private$C) <- colnames(responses)
         self$setVisualization()

R/PLNfamily-class.R

@@ -64,16 +64,22 @@ PLNfamily <-
       ## %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
       ## Post treatment --------------------
       #' @description Update fields after optimization
-      #' @param config a list for controlling the post-treatment.
-      postTreatment = function(config) {
-        nullModel <- nullModelPoisson(self$responses, self$covariates, self$offsets, self$weights)
+      #' @param config_post a list for controlling the post-treatments (optional bootstrap, jackknife, R2, etc.).
+      #' @param config_optim a list for controlling the optimization parameters used during post_treatments
+      postTreatment = function(config_post, config_optim) {
+        if (config_post$rsquared) {
+          nullModel <- nullModelPoisson(self$responses, self$covariates, self$offsets, self$weights)
+        } else {
+          nullModel <- NULL
+        }
         for (model in self$models)
           model$postTreatment(
             self$responses,
             self$covariates,
             self$offsets,
             self$weights,
-            config,
+            config_post=config_post,
+            config_optim=config_optim,
             nullModel = nullModel
         )
       },