feat: add subsampling to large data sets

.Rbuildignore

@@ -23,3 +23,4 @@
 .vscode
 ^kaggle$
 ^oml_cache$
+^vignettes$

.gitignore

@@ -98,3 +98,4 @@ rsconnect/
 .Rprofile
 kaggle/
 README.html
+inst/doc

DESCRIPTION

@@ -2,14 +2,14 @@ Package: mlr3automl
 Title: AutoML extention for 'mlr3'
 Version: 0.0.1
 Authors@R: c(
-    person("Damir", "Pulatov", , "[email protected]", role = c("cre", "aut")),
-    person("Marc", "Becker", , "[email protected]", role = "aut",
-           comment = c(ORCID = "0000-0002-8115-0400")),
+    person("Marc", "Becker", , "[email protected]", role = c("cre", "aut"),
+        comment = c(ORCID = "0000-0002-8115-0400")),
+    person("Damir", "Pulatov", , "[email protected]", role = "aut"),
     person("Baisu", "Zhou", , "[email protected]", role = "aut")
   )
 Description: Flexible AutoML system for the 'mlr3' ecosystem.
 License: LGPL-3
-URL: https://github.com/mlr-org/mlr3automl
+URL: https://github.com/mlr-org/mlr3automl https://mlr3automl.mlr-org.com
 BugReports: https://github.com/mlr-org/mlr3automl/issues
 Depends:
     mlr3 (>= 1.2.0.9000),
@@ -35,29 +35,32 @@ Suggests:
     fastai,
     glmnet,
     kknn,
+    knitr,
     lgr,
     lightgbm,
     MASS,
     mirai,
     mlr3extralearners,
     mlr3torch (>= 0.3.0.9000),
     mlr3viz,
+    quarto,
     ranger,
     redux,
     reticulate,
+    rmarkdown,
     rpart,
     testthat (>= 3.0.0),
     xgboost
 Remotes:
     catboost/catboost/catboost/R-package,
     eagerai/fastai,
+    mlr-org/bbotk,
     mlr-org/mlr3,
     mlr-org/mlr3extralearners,
     mlr-org/mlr3learners,
     mlr-org/mlr3mbo@so_config_6,
     mlr-org/mlr3torch,
     mlr-org/mlr3tuning,
-    mlr-org/bbotk,
     mlr-org/rush
 Config/testthat/edition: 3
 Config/testthat/parallel: false

R/Auto.R

@@ -3,7 +3,7 @@
 #' @description
 #' This class is the base class for all autos.
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @template param_id
 #' @template param_task
@@ -14,6 +14,8 @@
 #' @template param_large_data_set
 #' @template param_size
 #' @template param_devices
+#' @template param_pv
+#' @template param_graph
 #'
 #' @export
 Auto = R6Class("Auto",
@@ -80,7 +82,7 @@ Auto = R6Class("Auto",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       stop("Abstract")
     },
 
@@ -154,6 +156,12 @@ Auto = R6Class("Auto",
     #' Get the search space for the learner.
     search_space = function(task) {
       private$.search_space
+    },
+
+    #' @description
+    #' Modify the graph for the final model.
+    final_graph = function(graph, task, pv) {
+      graph
     }
   ),
 

R/AutoCatboost.R

@@ -1,6 +1,6 @@
 #' @title Catboost Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R Auto.R
 #'
 #' @description
 #' Catboost auto.
@@ -10,7 +10,12 @@
 #' @template param_measure
 #' @template param_n_threads
 #' @template param_timeout
+#' @template param_memory_limit
+#' @template param_large_data_set
+#' @template param_size
 #' @template param_devices
+#' @template param_pv
+#' @template param_graph
 #'
 #' @export
 AutoCatboost = R6Class("AutoCatboost",
@@ -31,7 +36,7 @@ AutoCatboost = R6Class("AutoCatboost",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)
@@ -52,10 +57,17 @@ AutoCatboost = R6Class("AutoCatboost",
         task_type = task_type)
       set_threads(learner, n_threads)
 
-      po("removeconstants", id = "catboost_removeconstants") %>>%
+
+      graph = po("removeconstants", id = "catboost_removeconstants") %>>%
         po("colapply", id = "catboost_colapply", applicator = as.numeric, affect_columns = selector_type("integer")) %>>%
         po("removeconstants", id = "catboost_post_removeconstants") %>>%
         learner
+
+      if (task$nrow * task$ncol > pv$large_data_size) {
+        graph = po("subsample", frac = 0.25, stratify = inherits(task, "TaskClassif"), use_groups = FALSE, id = "catboost_subsample") %>>% graph
+      }
+
+      graph
     },
 
     #' @description
@@ -115,6 +127,14 @@ AutoCatboost = R6Class("AutoCatboost",
           "merror" # default
         )
       }
+    },
+
+    #' @description
+    #' Modify the graph for the final model.
+    final_graph = function(graph, task, pv) {
+      if (task$nrow * task$ncol > pv$large_data_size) {
+        graph$param_set$set_values(catboost_subsample.frac = 1)
+      }
     }
   ),
 

R/AutoExtraTrees.R

@@ -1,17 +1,21 @@
 #' @title Extra Trees Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @description
 #' Extra Trees auto.
 #'
 #' @template param_id
-#' @template param_n_threads
-#' @template param_timeout
 #' @template param_task
 #' @template param_measure
+#' @template param_n_threads
+#' @template param_timeout
+#' @template param_memory_limit
+#' @template param_large_data_set
 #' @template param_size
 #' @template param_devices
+#' @template param_pv
+#' @template param_graph
 #'
 #' @export
 AutoExtraTrees = R6Class("AutoExtraTrees",
@@ -37,7 +41,7 @@ AutoExtraTrees = R6Class("AutoExtraTrees",
     #' @param measure ([mlr3::Measure]).
     #' @param n_threads (`numeric(1)`).
     #' @param timeout (`numeric(1)`).
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)
@@ -53,13 +57,19 @@ AutoExtraTrees = R6Class("AutoExtraTrees",
         sample.fraction = 1)
       set_threads(learner, n_threads)
 
-      po("removeconstants", id = "extra_trees_removeconstants") %>>%
+      graph = po("removeconstants", id = "extra_trees_removeconstants") %>>%
         po("imputeoor", id = "extra_trees_imputeoor") %>>%
         po("fixfactors", id = "extra_trees_fixfactors") %>>%
         po("imputesample", affect_columns = selector_type(c("factor", "ordered")), id = "extra_trees_imputesample") %>>%
         po("collapsefactors", target_level_count = 40, id = "extra_trees_collapse") %>>%
         po("removeconstants", id = "extra_trees_post_removeconstants") %>>%
         learner
+
+      if (task$nrow * task$ncol > pv$large_data_size) {
+        graph = po("subsample", frac = 0.25, stratify = inherits(task, "TaskClassif"), use_groups = FALSE, id = "extra_trees_subsample") %>>% graph
+      }
+
+      graph
     },
 
     #' @description
@@ -70,6 +80,14 @@ AutoExtraTrees = R6Class("AutoExtraTrees",
       num_trees = 100
       tree_size_bytes = task$nrow / 60000 * 1e6
       ceiling((tree_size_bytes * num_trees) / 1e6)
+    },
+
+    #' @description
+    #' Modify the graph for the final model.
+    final_graph = function(graph, task, pv) {
+      if (task$nrow * task$ncol > pv$large_data_size) {
+        graph$param_set$set_values(extra_trees_subsample.frac = 1)
+      }
     }
   )
 )

R/AutoFTTransformer.R

@@ -1,6 +1,6 @@
 #' @title FTTransformer Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @description
 #' FTTransformer auto.
@@ -11,6 +11,7 @@
 #' @template param_n_threads
 #' @template param_timeout
 #' @template param_devices
+#' @template param_pv
 #'
 #' @export
 AutoFTTransformer = R6Class("AutoFTTransformer",
@@ -31,7 +32,7 @@ AutoFTTransformer = R6Class("AutoFTTransformer",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)

R/AutoFastai.R

@@ -1,6 +1,6 @@
 #' @title Fastai Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @description
 #' Fastai auto.
@@ -13,6 +13,7 @@
 #' @template param_memory_limit
 #' @template param_large_data_set
 #' @template param_devices
+#' @template param_pv
 #'
 #' @export
 AutoFastai = R6Class("AutoFastai",
@@ -49,7 +50,7 @@ AutoFastai = R6Class("AutoFastai",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)

R/AutoGlmnet.R

@@ -1,6 +1,6 @@
 #' @title Glmnet Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @description
 #' Glmnet auto.
@@ -11,6 +11,7 @@
 #' @template param_n_threads
 #' @template param_timeout
 #' @template param_devices
+#' @template param_pv
 #'
 #' @export
 AutoGlmnet = R6Class("AutoGlmnet",
@@ -30,7 +31,7 @@ AutoGlmnet = R6Class("AutoGlmnet",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)

R/AutoKknn.R

@@ -1,6 +1,6 @@
 #' @title Kknn Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @description
 #' Kknn auto.
@@ -11,6 +11,7 @@
 #' @template param_n_threads
 #' @template param_timeout
 #' @template param_devices
+#' @template param_pv
 #'
 #' @export
 AutoKknn = R6Class("AutoKknn",
@@ -30,7 +31,7 @@ AutoKknn = R6Class("AutoKknn",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)

R/AutoLda.R

@@ -1,6 +1,6 @@
 #' @title Lda Auto
 #'
-#' @include mlr_auto.R
+#' @include mlr_auto.R Auto.R
 #'
 #' @description
 #' Lda auto.
@@ -14,6 +14,7 @@
 #' @template param_n_threads
 #' @template param_timeout
 #' @template param_devices
+#' @template param_pv
 #'
 #' @export
 AutoLda = R6Class("AutoLda",
@@ -33,7 +34,7 @@ AutoLda = R6Class("AutoLda",
 
     #' @description
     #' Create the graph for the auto.
-    graph = function(task, measure, n_threads, timeout, devices) {
+    graph = function(task, measure, n_threads, timeout, devices, pv) {
       assert_task(task)
       assert_measure(measure)
       assert_count(n_threads)