Batch size not required

DESCRIPTION

@@ -27,3 +27,4 @@ Suggests:
     ggplot2,
     future, future.apply
 VignetteBuilder: knitr
+Remotes: mlr-org/mlr3torch@dont-require-batch-size

man/batch_sampler_stratified.Rd

@@ -42,24 +42,20 @@ count_tab/sum(count_tab) # ~10% R, ~90% M.
 
 ## Not a typical use case, but shows class distribution in each batch.
 sonar_list <- list(task=sonar_task)
-batch_sampler_class <- mlr3torchAUM::batch_sampler_stratified(
-  min_samples_per_stratum = 1)
+batch_sampler_class <- mlr3torchAUM::batch_sampler_stratified(min_samples_per_stratum = 1)
 batch_sampler_instance <- batch_sampler_class(sonar_list)
 ## In the counts below there is 1 R per batch/column.
 sapply(batch_sampler_instance$batch_list, function(i)table(Class_vec[i]))
 
 ## Whereas some batches have 0 R with random:
-random_sampler_class <- mlr3torchAUM::batch_sampler_random(
-  batch_size = 10)
+random_sampler_class <- mlr3torchAUM::batch_sampler_random(batch_size = 10)
 random_sampler_instance <- random_sampler_class(sonar_list)
 sapply(random_sampler_instance$batch_list, function(i)table(Class_vec[i]))
 
 ## Typical use case below is as the batch_sampler param in a TorchLearner.
 if(torch::torch_is_installed() && requireNamespace("mlr3torch")){
   L <- mlr3torch::LearnerTorchMLP$new(task_type="classif")
-  L$param_set$set_values(
-    epochs=1, batch_size=10, seed=1,
-    batch_sampler=batch_sampler_class)
+  L$param_set$set_values(epochs=1, shuffle=NULL, batch_sampler=batch_sampler_class)
   L$train(sonar_task)
 }
 

vignettes/Typical_usage.Rmd

@@ -115,7 +115,7 @@ if(torch_available){
     L$predict_type <- "prob"
     L$callbacks <- mlr3torch::t_clbk("history")
     L$param_set$set_values(
-      epochs=20, batch_size=batch_size, seed=1,
+      epochs=20, shuffle=NULL, seed=1,
       measures_valid=measure_list,
       measures_train=measure_list,
       batch_sampler=batch_sampler_list[[batching]])