Runs a benchmark on arbitrary combinations of tasks (Task), learners (Learner), and resampling strategies (Resampling), possibly in parallel.

## Usage

benchmark(
design,
store_models = FALSE,
store_backends = TRUE,
encapsulate = NA_character_,
allow_hotstart = FALSE,
)

## Arguments

design

(data.frame())
Data frame (or data.table::data.table()) with three columns: "task", "learner", and "resampling". Each row defines a resampling by providing a Task, Learner and an instantiated Resampling strategy. The helper function benchmark_grid() can assist in generating an exhaustive design (see examples) and instantiate the Resamplings per Task.

store_models

(logical(1))
Store the fitted model in the resulting object= Set to TRUE if you want to further analyse the models or want to extract information like variable importance.

store_backends

(logical(1))
Keep the DataBackend of the Task in the ResampleResult? Set to TRUE if your performance measures require a Task, or to analyse results more conveniently. Set to FALSE to reduce the file size and memory footprint after serialization. The current default is TRUE, but this eventually will be changed in a future release.

encapsulate

(character(1))
If not NA, enables encapsulation by setting the field Learner$encapsulate to one of the supported values: "none" (disable encapsulation), "evaluate" (execute via evaluate) and "callr" (start in external session via callr). If NA, encapsulation is not changed, i.e. the settings of the individual learner are active. Additionally, if encapsulation is set to "evaluate" or "callr", the fallback learner is set to the featureless learner if the learner does not already have a fallback configured. allow_hotstart (logical(1)) Determines if learner(s) are hot started with trained models in $hotstart_stack. See also HotstartStack.

clone

(character())
Select the input objects to be cloned before proceeding by providing a set with possible values "task", "learner" and "resampling" for Task, Learner and Resampling, respectively. Per default, all input objects are cloned.

## Note

The fitted models are discarded after the predictions have been scored in order to reduce memory consumption. If you need access to the models for later analysis, set store_models to TRUE.

## Predict Sets

If you want to compare the performance of a learner on the training with the performance on the test set, you have to configure the Learner to predict on multiple sets by setting the field predict_sets to c("train", "test") (default is "test"). Each set yields a separate Prediction object during resampling. In the next step, you have to configure the measures to operate on the respective Prediction object:

m1 = msr("classif.ce", id = "ce.train", predict_sets = "train")
m2 = msr("classif.ce", id = "ce.test", predict_sets = "test")

The (list of) created measures can finally be passed to $aggregate() or $score().

## Parallelization

This function can be parallelized with the future package. One job is one resampling iteration, and all jobs are send to an apply function from future.apply in a single batch. To select a parallel backend, use future::plan().

## Progress Bars

This function supports progress bars via the package progressr. Simply wrap the function call in progressr::with_progress() to enable them. Alternatively, call progressr::handlers() with global = TRUE to enable progress bars globally. We recommend the progress package as backend which can be enabled with progressr::handlers("progress").

## Logging

The mlr3 uses the lgr package for logging. lgr supports multiple log levels which can be queried with getOption("lgr.log_levels").

To suppress output and reduce verbosity, you can lower the log from the default level "info" to "warn":

lgr::get_logger("mlr3")$set_threshold("warn") To get additional log output for debugging, increase the log level to "debug" or "trace": lgr::get_logger("mlr3")$set_threshold("debug")

To log to a file or a data base, see the documentation of lgr::lgr-package.

Other benchmark: BenchmarkResult, benchmark_grid()

## Examples

# benchmarking with benchmark_grid()
learners = lapply(c("classif.featureless", "classif.rpart"), lrn)
resamplings = rsmp("cv", folds = 3)

print(design)
#> 1: penguins classif.featureless         cv
#> 2: penguins       classif.rpart         cv
#> 3:    sonar classif.featureless         cv
#> 4:    sonar       classif.rpart         cv

set.seed(123)
bmr = benchmark(design)

## Data of all resamplings
#>                            learner         resampling iteration
#> 1: <LearnerClassifFeatureless[38]> <ResamplingCV[20]>         1
#> 2: <LearnerClassifFeatureless[38]> <ResamplingCV[20]>         2
#> 3: <LearnerClassifFeatureless[38]> <ResamplingCV[20]>         3
#> 4:       <LearnerClassifRpart[38]> <ResamplingCV[20]>         1
#> 5:       <LearnerClassifRpart[38]> <ResamplingCV[20]>         2
#> 6:       <LearnerClassifRpart[38]> <ResamplingCV[20]>         3
#>                 prediction
#> 1: <PredictionClassif[20]>
#> 2: <PredictionClassif[20]>
#> 3: <PredictionClassif[20]>
#> 4: <PredictionClassif[20]>
#> 5: <PredictionClassif[20]>
#> 6: <PredictionClassif[20]>

## Aggregated performance values
aggr = bmr$aggregate() print(aggr) #> nr task_id learner_id resampling_id iters classif.ce #> 1: 1 penguins classif.featureless cv 3 0.55822527 #> 2: 2 penguins classif.rpart cv 3 0.06112382 #> 3: 3 sonar classif.featureless cv 3 0.46611456 #> 4: 4 sonar classif.rpart cv 3 0.25458937 #> Hidden columns: resample_result ## Extract predictions of first resampling result rr = aggr$resample_result[[1]]
as.data.table(rr$prediction()) #> row_ids truth response #> 1: 1 Adelie Adelie #> 2: 2 Adelie Adelie #> 3: 6 Adelie Adelie #> 4: 11 Adelie Adelie #> 5: 17 Adelie Adelie #> --- #> 340: 328 Chinstrap Adelie #> 341: 331 Chinstrap Adelie #> 342: 333 Chinstrap Adelie #> 343: 336 Chinstrap Adelie #> 344: 344 Chinstrap Adelie # Benchmarking with a custom design: # - fit classif.featureless on penguins with a 3-fold CV # - fit classif.rpart on sonar using a holdout tasks = list(tsk("penguins"), tsk("sonar")) learners = list(lrn("classif.featureless"), lrn("classif.rpart")) resamplings = list(rsmp("cv", folds = 3), rsmp("holdout")) design = data.table::data.table( task = tasks, learner = learners, resampling = resamplings ) ## Instantiate resamplings design$resampling = Map(
function(task, resampling) resampling$clone()$instantiate(task),
task = design$task, resampling = design$resampling
)

## Run benchmark
bmr = benchmark(design)
print(bmr)
#> <BenchmarkResult> of 4 rows with 2 resampling runs
#>  nr  task_id          learner_id resampling_id iters warnings errors
#>   1 penguins classif.featureless            cv     3        0      0
#>   2    sonar       classif.rpart       holdout     1        0      0

## Get the training set of the 2nd iteration of the featureless learner on penguins
rr = bmr$aggregate()[learner_id == "classif.featureless"]$resample_result[[1]]
rr$resampling$train_set(2)
#>   [1]   5   7   8   9  12  13  17  19  22  25  28  35  36  40  46  48  49  50
#>  [19]  52  53  54  60  61  62  63  67  69  72  73  74  75  76  78  81  84  85
#>  [37]  88  92  97 101 103 104 109 110 114 119 122 127 129 130 131 136 145 147
#>  [55] 156 160 162 163 166 170 172 173 177 179 180 184 185 188 190 191 193 194
#>  [73] 205 212 213 217 218 221 228 229 233 234 238 239 245 250 252 255 256 258
#>  [91] 259 263 264 267 271 278 281 282 287 291 295 296 297 300 302 307 309 317
#> [109] 319 321 325 327 331 337 341   3   4  10  11  15  23  24  31  32  38  39
#> [127]  41  42  43  45  56  57  58  64  68  77  79  87  93  99 100 102 111 112
#> [145] 113 115 118 120 124 126 128 132 133 134 135 137 142 143 146 148 149 151
#> [163] 153 154 155 159 161 164 169 171 174 182 186 189 195 196 197 199 200 203
#> [181] 204 207 210 215 219 220 223 224 226 227 230 240 242 243 246 247 251 253
#> [199] 261 266 269 273 274 275 276 277 279 284 286 288 289 290 292 293 301 306
#> [217] 308 313 314 315 316 318 323 330 335 336 338 343 344