rpart: Multiclass Classification

# nolint start
library(mlexperiments)

See https://github.com/kapsner/mlexperiments/blob/main/R/learner_rpart.R for implementation details.

Preprocessing

Import and Prepare Data

library(mlbench)
data("DNA")
dataset <- DNA |>
  data.table::as.data.table() |>
  na.omit()

feature_cols <- colnames(dataset)[1:180]
target_col <- "Class"

General Configurations

seed <- 123
if (isTRUE(as.logical(Sys.getenv("_R_CHECK_LIMIT_CORES_")))) {
  # on cran
  ncores <- 2L
} else {
  ncores <- ifelse(
    test = parallel::detectCores() > 4,
    yes = 4L,
    no = ifelse(
      test = parallel::detectCores() < 2L,
      yes = 1L,
      no = parallel::detectCores()
    )
  )
}
options("mlexperiments.bayesian.max_init" = 10L)

Generate Training- and Test Data

data_split <- splitTools::partition(
  y = dataset[, get(target_col)],
  p = c(train = 0.7, test = 0.3),
  type = "stratified",
  seed = seed
)

train_x <- model.matrix(
  ~ -1 + .,
  dataset[data_split$train, .SD, .SDcols = feature_cols]
)
train_y <- dataset[data_split$train, get(target_col)]


test_x <- model.matrix(
  ~ -1 + .,
  dataset[data_split$test, .SD, .SDcols = feature_cols]
)
test_y <- dataset[data_split$test, get(target_col)]

Generate Training Data Folds

fold_list <- splitTools::create_folds(
  y = train_y,
  k = 3,
  type = "stratified",
  seed = seed
)

Experiments

Prepare Experiments

# required learner arguments, not optimized
learner_args <- list(method = "class")

# set arguments for predict function and performance metric,
# required for mlexperiments::MLCrossValidation and
# mlexperiments::MLNestedCV
predict_args <- list(type = "class")
performance_metric <- metric("bacc")
performance_metric_args <- NULL
return_models <- FALSE

# required for grid search and initialization of bayesian optimization
parameter_grid <- expand.grid(
  minsplit = seq(2L, 82L, 10L),
  cp = seq(0.01, 0.1, 0.01),
  maxdepth = seq(2L, 30L, 5L)
)
# reduce to a maximum of 10 rows
if (nrow(parameter_grid) > 10) {
  set.seed(123)
  sample_rows <- sample(seq_len(nrow(parameter_grid)), 10, FALSE)
  parameter_grid <- kdry::mlh_subset(parameter_grid, sample_rows)
}

# required for bayesian optimization
parameter_bounds <- list(
  minsplit = c(2L, 100L),
  cp = c(0.01, 0.1),
  maxdepth = c(2L, 30L)
)
optim_args <- list(
  iters.n = ncores,
  kappa = 3.5,
  acq = "ucb"
)

Hyperparameter Tuning

tuner <- mlexperiments::MLTuneParameters$new(
  learner = LearnerRpart$new(),
  strategy = "grid",
  ncores = ncores,
  seed = seed
)

tuner$parameter_grid <- parameter_grid
tuner$learner_args <- learner_args
tuner$split_type <- "stratified"

tuner$set_data(
  x = train_x,
  y = train_y
)

tuner_results_grid <- tuner$execute(k = 3)
#> 
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================>-----------------------------------------------------------------------------] 2/10 ( 20%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================>-------------------------------------------------------------------] 3/10 ( 30%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================>----------------------------------------------------------] 4/10 ( 40%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================>------------------------------------------------] 5/10 ( 50%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=========================================================>--------------------------------------] 6/10 ( 60%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================================================================>-----------------------------] 7/10 ( 70%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================================================================>-------------------] 8/10 ( 80%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================================================================>----------] 9/10 ( 90%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================================================================] 10/10 (100%)                                                                                                                                  
#>  Classification: using 'classification error rate' as optimization metric.

head(tuner_results_grid)
#>    setting_id metric_optim_mean minsplit   cp maxdepth method
#> 1:          1        0.09465558        2 0.07       22  class
#> 2:          2        0.09465558       32 0.02       27  class
#> 3:          3        0.09465558       72 0.10        7  class
#> 4:          4        0.09465558       32 0.09       27  class
#> 5:          5        0.09465558       52 0.02       12  class
#> 6:          6        0.09465558        2 0.04        7  class

Bayesian Optimization

tuner <- mlexperiments::MLTuneParameters$new(
  learner = LearnerRpart$new(),
  strategy = "bayesian",
  ncores = ncores,
  seed = seed
)

tuner$parameter_grid <- parameter_grid
tuner$parameter_bounds <- parameter_bounds

tuner$learner_args <- learner_args
tuner$optim_args <- optim_args

tuner$split_type <- "stratified"

tuner$set_data(
  x = train_x,
  y = train_y
)

tuner_results_bayesian <- tuner$execute(k = 3)
#> 
#> Registering parallel backend using 4 cores.

head(tuner_results_bayesian)
#>    Epoch setting_id minsplit   cp maxdepth gpUtility acqOptimum inBounds Elapsed       Score metric_optim_mean errorMessage method
#> 1:     0          1        2 0.07       22        NA      FALSE     TRUE   2.108 -0.09465558        0.09465558           NA  class
#> 2:     0          2       32 0.02       27        NA      FALSE     TRUE   2.122 -0.09465558        0.09465558           NA  class
#> 3:     0          3       72 0.10        7        NA      FALSE     TRUE   2.025 -0.09465558        0.09465558           NA  class
#> 4:     0          4       32 0.09       27        NA      FALSE     TRUE   2.258 -0.09465558        0.09465558           NA  class
#> 5:     0          5       52 0.02       12        NA      FALSE     TRUE   2.030 -0.09465558        0.09465558           NA  class
#> 6:     0          6        2 0.04        7        NA      FALSE     TRUE   2.099 -0.09465558        0.09465558           NA  class

k-Fold Cross Validation

validator <- mlexperiments::MLCrossValidation$new(
  learner = LearnerRpart$new(),
  fold_list = fold_list,
  ncores = ncores,
  seed = seed
)

validator$learner_args <- tuner$results$best.setting[-1]

validator$predict_args <- predict_args
validator$performance_metric <- performance_metric
validator$performance_metric_args <- performance_metric_args
validator$return_models <- return_models

validator$set_data(
  x = train_x,
  y = train_y
)

validator_results <- validator$execute()
#> 
#> CV fold: Fold1
#> 
#> CV fold: Fold2
#> CV progress [====================================================================>-----------------------------------] 2/3 ( 67%)
#> 
#> CV fold: Fold3
#> CV progress [========================================================================================================] 3/3 (100%)
#>                                                                                                                                   

head(validator_results)
#>     fold performance minsplit   cp maxdepth method
#> 1: Fold1   0.8950174        2 0.07       22  class
#> 2: Fold2   0.8978974        2 0.07       22  class
#> 3: Fold3   0.8917513        2 0.07       22  class

Nested Cross Validation

validator <- mlexperiments::MLNestedCV$new(
  learner = LearnerRpart$new(),
  strategy = "grid",
  fold_list = fold_list,
  k_tuning = 3L,
  ncores = ncores,
  seed = seed
)

validator$parameter_grid <- parameter_grid
validator$learner_args <- learner_args
validator$split_type <- "stratified"

validator$predict_args <- predict_args
validator$performance_metric <- performance_metric
validator$performance_metric_args <- performance_metric_args
validator$return_models <- return_models

validator$set_data(
  x = train_x,
  y = train_y
)

validator_results <- validator$execute()
#> 
#> CV fold: Fold1
#> 
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================>-----------------------------------------------------------------------------] 2/10 ( 20%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================>-------------------------------------------------------------------] 3/10 ( 30%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================>----------------------------------------------------------] 4/10 ( 40%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================>------------------------------------------------] 5/10 ( 50%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=========================================================>--------------------------------------] 6/10 ( 60%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================================================================>-----------------------------] 7/10 ( 70%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================================================================>-------------------] 8/10 ( 80%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================================================================>----------] 9/10 ( 90%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================================================================] 10/10 (100%)                                                                                                                                  
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> CV fold: Fold2
#> CV progress [====================================================================>-----------------------------------] 2/3 ( 67%)
#> 
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================>-----------------------------------------------------------------------------] 2/10 ( 20%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================>-------------------------------------------------------------------] 3/10 ( 30%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================>----------------------------------------------------------] 4/10 ( 40%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================>------------------------------------------------] 5/10 ( 50%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=========================================================>--------------------------------------] 6/10 ( 60%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================================================================>-----------------------------] 7/10 ( 70%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================================================================>-------------------] 8/10 ( 80%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================================================================>----------] 9/10 ( 90%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================================================================] 10/10 (100%)                                                                                                                                  
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> CV fold: Fold3
#> CV progress [========================================================================================================] 3/3 (100%)
#>                                                                                                                                   
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================>-----------------------------------------------------------------------------] 2/10 ( 20%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================>-------------------------------------------------------------------] 3/10 ( 30%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================>----------------------------------------------------------] 4/10 ( 40%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================>------------------------------------------------] 5/10 ( 50%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=========================================================>--------------------------------------] 6/10 ( 60%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [==================================================================>-----------------------------] 7/10 ( 70%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [============================================================================>-------------------] 8/10 ( 80%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [=====================================================================================>----------] 9/10 ( 90%)
#>  Classification: using 'classification error rate' as optimization metric.
#> 
#> Parameter settings [===============================================================================================] 10/10 (100%)                                                                                                                                  
#>  Classification: using 'classification error rate' as optimization metric.

head(validator_results)
#>     fold performance minsplit   cp maxdepth method
#> 1: Fold1   0.8950174        2 0.07       22  class
#> 2: Fold2   0.8978974        2 0.07       22  class
#> 3: Fold3   0.8917513        2 0.07       22  class

Inner Bayesian Optimization

validator <- mlexperiments::MLNestedCV$new(
  learner = LearnerRpart$new(),
  strategy = "bayesian",
  fold_list = fold_list,
  k_tuning = 3L,
  ncores = ncores,
  seed = seed
)

validator$parameter_grid <- parameter_grid
validator$learner_args <- learner_args
validator$split_type <- "stratified"


validator$parameter_bounds <- parameter_bounds
validator$optim_args <- optim_args

validator$predict_args <- predict_args
validator$performance_metric <- performance_metric
validator$performance_metric_args <- performance_metric_args
validator$return_models <- return_models

validator$set_data(
  x = train_x,
  y = train_y
)

validator_results <- validator$execute()
#> 
#> CV fold: Fold1
#> 
#> Registering parallel backend using 4 cores.
#> 
#> CV fold: Fold2
#> CV progress [====================================================================>-----------------------------------] 2/3 ( 67%)
#> 
#> Registering parallel backend using 4 cores.
#> 
#> CV fold: Fold3
#> CV progress [========================================================================================================] 3/3 (100%)
#>                                                                                                                                   
#> Registering parallel backend using 4 cores.

head(validator_results)
#>     fold performance minsplit   cp maxdepth method
#> 1: Fold1   0.8950174        2 0.07       22  class
#> 2: Fold2   0.8978974        2 0.07       22  class
#> 3: Fold3   0.8917513        2 0.07       22  class

Appendix I: Grid-Search with Target Weigths

Here, rpart’s weights-argument is used to rescale the case-weights during the training.

# define the target weights
y_weights <- ifelse(train_y == "n", 0.8, ifelse(train_y == "ei", 1.2, 1))
head(y_weights)
#> [1] 1.2 1.2 0.0 0.8 0.8 0.0
tuner_w_weights <- mlexperiments::MLTuneParameters$new(
  learner = LearnerRpart$new(),
  strategy = "grid",
  ncores = ncores,
  seed = seed
)

tuner_w_weights$parameter_grid <- parameter_grid
tuner_w_weights$learner_args <- c(
  learner_args,
  list(case_weights = y_weights)
)
tuner_w_weights$split_type <- "stratified"

tuner_w_weights$set_data(
  x = train_x,
  y = train_y
)

tuner_results_grid <- tuner_w_weights$execute(k = 3)
#> 
#> Parameter settings [============================>-------------------------------------------------------------------] 3/10 ( 30%)
#> Parameter settings [=====================================>----------------------------------------------------------] 4/10 ( 40%)
#> Parameter settings [===============================================>------------------------------------------------] 5/10 ( 50%)
#> Parameter settings [=========================================================>--------------------------------------] 6/10 ( 60%)
#> Parameter settings [==================================================================>-----------------------------] 7/10 ( 70%)
#> Parameter settings [============================================================================>-------------------] 8/10 ( 80%)
#> Parameter settings [=====================================================================================>----------] 9/10 ( 90%)
#> Parameter settings [===============================================================================================] 10/10 (100%)                                                                                                                                  

head(tuner_results_grid)
#>    setting_id metric_optim_mean minsplit    cp maxdepth method
#>         <int>             <num>    <int> <num>    <int> <char>
#> 1:          1         0.1062916        2  0.07       22  class
#> 2:          2         0.1062916       32  0.02       27  class
#> 3:          3         0.1062916       72  0.10        7  class
#> 4:          4         0.1062916       32  0.09       27  class
#> 5:          5         0.1062916       52  0.02       12  class
#> 6:          6         0.1062916        2  0.04        7  class

Appendix II: k-Fold Cross Validation with Target Weigths

validator <- mlexperiments::MLCrossValidation$new(
  learner = LearnerRpart$new(),
  fold_list = fold_list,
  ncores = ncores,
  seed = seed
)

# append the optimized setting from above with the newly created weights
validator$learner_args <- c(
  tuner$results$best.setting[-1],
  list("case_weights" = y_weights)
)

validator$predict_args <- predict_args
validator$performance_metric <- performance_metric
validator$performance_metric_args <- performance_metric_args
validator$return_models <- return_models

validator$set_data(
  x = train_x,
  y = train_y
)

validator_results <- validator$execute()
#> 
#> CV fold: Fold1
#> 
#> CV fold: Fold2
#> 
#> CV fold: Fold3
#> CV progress [========================================================================================================] 3/3 (100%)
#>                                                                                                                                   

head(validator_results)
#>      fold performance minsplit    cp maxdepth method
#>    <char>       <num>    <num> <num>    <num> <char>
#> 1:  Fold1   0.8812005        2  0.07       22  class
#> 2:  Fold2   0.9129256        2  0.07       22  class
#> 3:  Fold3   0.8800668        2  0.07       22  class