The TuningInstanceSingleCrit specifies a tuning problem for Tuners.
The function ti() creates a TuningInstanceSingleCrit and the function tune() creates an instance internally.
Details
The instance contains an ObjectiveTuning object that encodes the black box objective function a Tuner has to optimize.
The instance allows the basic operations of querying the objective at design points ($eval_batch()).
This operation is usually done by the Tuner.
Evaluations of hyperparameter configurations are performed in batches by calling mlr3::benchmark() internally.
The evaluated hyperparameter configurations are stored in the Archive ($archive).
Before a batch is evaluated, the bbotk::Terminator is queried for the remaining budget.
If the available budget is exhausted, an exception is raised, and no further evaluations can be performed from this point on.
The tuner is also supposed to store its final result, consisting of a selected hyperparameter configuration and associated estimated performance values, by calling the method instance$assign_result.
Default Measures
If no measure is passed, the default measure is used. The default measure depends on the task type.
| Task | Default Measure | Package |
"classif" | "classif.ce" | mlr3 |
"regr" | "regr.mse" | mlr3 |
"surv" | "surv.cindex" | mlr3proba |
"dens" | "dens.logloss" | mlr3proba |
"classif_st" | "classif.ce" | mlr3spatial |
"regr_st" | "regr.mse" | mlr3spatial |
"clust" | "clust.dunn" | mlr3cluster |
Resources
There are several sections about hyperparameter optimization in the mlr3book.
Getting started with hyperparameter optimization.
Tune a simple classification tree on the Sonar data set.
Learn about tuning spaces.
The gallery features a collection of case studies and demos about optimization.
Learn more advanced methods with the practical tuning series.
Simultaneously optimize hyperparameters and use early stopping with XGBoost.
Make us of proven search space.
Learn about hotstarting models.
Run the default hyperparameter configuration of learners as a baseline.
Extension Packages
mlr3tuning is extended by the following packages.
mlr3tuningspaces is a collection of search spaces from scientific articles for commonly used learners.
mlr3hyperband adds the Hyperband and Successive Halving algorithm.
mlr3mbo adds Bayesian optimization methods.
Analysis
For analyzing the tuning results, it is recommended to pass the ArchiveTuning to as.data.table().
The returned data table is joined with the benchmark result which adds the mlr3::ResampleResult for each hyperparameter evaluation.
The archive provides various getters (e.g. $learners()) to ease the access.
All getters extract by position (i) or unique hash (uhash).
For a complete list of all getters see the methods section.
The benchmark result ($benchmark_result) allows to score the hyperparameter configurations again on a different measure.
Alternatively, measures can be supplied to as.data.table().
The mlr3viz package provides visualizations for tuning results.
Super classes
bbotk::OptimInstance -> bbotk::OptimInstanceSingleCrit -> TuningInstanceSingleCrit
Active bindings
result_learner_param_vals(
list())
Param values for the optimal learner call.
Methods
Method new()
Creates a new instance of this R6 class.
Usage
TuningInstanceSingleCrit$new(
task,
learner,
resampling,
measure = NULL,
terminator,
search_space = NULL,
store_benchmark_result = TRUE,
store_models = FALSE,
check_values = FALSE,
allow_hotstart = FALSE,
keep_hotstart_stack = FALSE,
evaluate_default = FALSE,
callbacks = list()
)Arguments
task(mlr3::Task)
Task to operate on.learner(mlr3::Learner)
Learner to tune.resampling(mlr3::Resampling)
Resampling that is used to evaluate the performance of the hyperparameter configurations. Uninstantiated resamplings are instantiated during construction so that all configurations are evaluated on the same data splits. Already instantiated resamplings are kept unchanged. Specialized Tuner change the resampling e.g. to evaluate a hyperparameter configuration on different data splits. This field, however, always returns the resampling passed in construction.measure(mlr3::Measure)
Measure to optimize. IfNULL, default measure is used.terminator(Terminator)
Stop criterion of the tuning process.search_space(paradox::ParamSet)
Hyperparameter search space. IfNULL(default), the search space is constructed from the TuneToken of the learner's parameter set (learner$param_set).store_benchmark_result(
logical(1))
IfTRUE(default), store resample result of evaluated hyperparameter configurations in archive as mlr3::BenchmarkResult.store_models(
logical(1))
IfTRUE, fitted models are stored in the benchmark result (archive$benchmark_result). Ifstore_benchmark_result = FALSE, models are only stored temporarily and not accessible after the tuning. This combination is needed for measures that require a model.check_values(
logical(1))
IfTRUE, hyperparameter values are checked before evaluation and performance scores after. IfFALSE(default), values are unchecked but computational overhead is reduced.allow_hotstart(
logical(1))
Allow to hotstart learners with previously fitted models. See also mlr3::HotstartStack. The learner must support hotstarting. Setsstore_models = TRUE.keep_hotstart_stack(
logical(1))
IfTRUE, mlr3::HotstartStack is kept in$objective$hotstart_stackafter tuning.evaluate_default(
logical(1))
IfTRUE, learner is evaluated with hyperparameters set to their default values at the start of the optimization.callbacks(list of CallbackTuning)
List of callbacks.
Method assign_result()
The Tuner object writes the best found point and estimated performance value here. For internal use.
Arguments
xdt(
data.table::data.table())
Hyperparameter values asdata.table::data.table(). Each row is one configuration. Contains values in the search space. Can contain additional columns for extra information.y(
numeric(1))
Optimal outcome.learner_param_vals(List of named
list()s)
Fixed parameter values of the learner that are neither part of the
Examples
# Hyperparameter optimization on the Palmer Penguins data set
task = tsk("penguins")
# Load learner and set search space
learner = lrn("classif.rpart",
cp = to_tune(1e-04, 1e-1, logscale = TRUE)
)
# Construct tuning instance
instance = ti(
task = task,
learner = learner,
resampling = rsmp("cv", folds = 3),
measures = msr("classif.ce"),
terminator = trm("evals", n_evals = 4)
)
# Choose optimization algorithm
tuner = tnr("random_search", batch_size = 2)
# Run tuning
tuner$optimize(instance)
#> cp learner_param_vals x_domain classif.ce
#> <num> <list> <list> <num>
#> 1: -7.564052 <list[2]> <list[1]> 0.0726926
# Set optimal hyperparameter configuration to learner
learner$param_set$values = instance$result_learner_param_vals
# Train the learner on the full data set
learner$train(task)
# Inspect all evaluated configurations
as.data.table(instance$archive)
#> cp classif.ce x_domain_cp runtime_learners timestamp
#> <num> <num> <num> <num> <POSc>
#> 1: -7.564052 0.0726926 0.0005187690 0.014 2024-03-06 08:51:13
#> 2: -7.420301 0.0726926 0.0005989688 0.018 2024-03-06 08:51:13
#> 3: -4.928764 0.0726926 0.0072354403 0.016 2024-03-06 08:51:13
#> 4: -7.684273 0.0726926 0.0004600050 0.015 2024-03-06 08:51:13
#> batch_nr warnings errors resample_result
#> <int> <int> <int> <list>
#> 1: 1 0 0 <ResampleResult>
#> 2: 1 0 0 <ResampleResult>
#> 3: 2 0 0 <ResampleResult>
#> 4: 2 0 0 <ResampleResult>