Subclass for grid search tuning.
Details
The grid is constructed as a Cartesian product over discretized values per parameter, see paradox::generate_design_grid().
If the learner supports hotstarting, the grid is sorted by the hotstart parameter (see also mlr3::HotstartStack).
If not, the points of the grid are evaluated in a random order.
Dictionary
This Tuner can be instantiated with the associated sugar function tnr():
tnr("grid_search")Control Parameters
resolutioninteger(1)
Resolution of the grid, seeparadox::generate_design_grid().param_resolutionsnamed
integer()
Resolution per parameter, named by parameter ID, seeparadox::generate_design_grid().batch_sizeinteger(1)
Maximum number of points to try in a batch.
Progress Bars
$optimize() supports progress bars via the package progressr
combined with a Terminator. Simply wrap the function in
progressr::with_progress() to enable them. We recommend to use package
progress as backend; enable with progressr::handlers("progress").
Parallelization
In order to support general termination criteria and parallelization, we
evaluate points in a batch-fashion of size batch_size. Larger batches mean
we can parallelize more, smaller batches imply a more fine-grained checking
of termination criteria. A batch contains of batch_size times resampling$iters jobs.
E.g., if you set a batch size of 10 points and do a 5-fold cross validation, you can
utilize up to 50 cores.
Parallelization is supported via package future (see mlr3::benchmark()'s
section on parallelization for more details).
Logging
All Tuners use a logger (as implemented in lgr) from package
bbotk.
Use lgr::get_logger("bbotk") to access and control the logger.
Optimizer
This Tuner is based on bbotk::OptimizerGridSearch which can be applied on any black box optimization problem. See also the documentation of bbotk.
Resources
book section on tuners.
mlr3hyperband extension package for the Hyperband algorithm.
Super class
mlr3tuning::Tuner -> TunerGridSearch
Examples
# Hyperparameter Optimization
# load learner and set search space
learner = lrn("classif.rpart",
cp = to_tune(1e-04, 1e-1, logscale = TRUE)
)
# run hyperparameter tuning on the Palmer Penguins data set
instance = tune(
method = "grid_search",
task = tsk("penguins"),
learner = learner,
resampling = rsmp("holdout"),
measure = msr("classif.ce"),
term_evals = 10
)
# best performing hyperparameter configuration
instance$result
#> cp learner_param_vals x_domain classif.ce
#> 1: -2.302585 <list[2]> <list[1]> 0.08695652
# all evaluated hyperparameter configuration
as.data.table(instance$archive)
#> cp classif.ce x_domain_cp runtime_learners timestamp
#> 1: -2.302585 0.08695652 0.1000000000 0.012 2022-12-22 07:40:18
#> 2: -6.907755 0.08695652 0.0010000000 0.011 2022-12-22 07:40:18
#> 3: -4.605170 0.08695652 0.0100000000 0.012 2022-12-22 07:40:18
#> 4: -3.837642 0.08695652 0.0215443469 0.013 2022-12-22 07:40:18
#> 5: -3.070113 0.08695652 0.0464158883 0.011 2022-12-22 07:40:18
#> 6: -8.442812 0.08695652 0.0002154435 0.011 2022-12-22 07:40:18
#> 7: -9.210340 0.08695652 0.0001000000 0.011 2022-12-22 07:40:18
#> 8: -5.372699 0.08695652 0.0046415888 0.011 2022-12-22 07:40:18
#> 9: -7.675284 0.08695652 0.0004641589 0.012 2022-12-22 07:40:18
#> 10: -6.140227 0.08695652 0.0021544347 0.010 2022-12-22 07:40:18
#> batch_nr warnings errors resample_result
#> 1: 1 0 0 <ResampleResult[21]>
#> 2: 2 0 0 <ResampleResult[21]>
#> 3: 3 0 0 <ResampleResult[21]>
#> 4: 4 0 0 <ResampleResult[21]>
#> 5: 5 0 0 <ResampleResult[21]>
#> 6: 6 0 0 <ResampleResult[21]>
#> 7: 7 0 0 <ResampleResult[21]>
#> 8: 8 0 0 <ResampleResult[21]>
#> 9: 9 0 0 <ResampleResult[21]>
#> 10: 10 0 0 <ResampleResult[21]>
# fit final model on complete data set
learner$param_set$values = instance$result_learner_param_vals
learner$train(tsk("penguins"))