Skip to contents

Hyperparameter Tuning with Covariance Matrix Adaptation Evolution Strategy

Source: R/TunerBatchCmaes.R
mlr_tuners_cmaes.Rd

Subclass for Covariance Matrix Adaptation Evolution Strategy (CMA-ES). Calls libcmaesr::cmaes() from package libcmaesr.

Source

Hansen N (2016). “The CMA Evolution Strategy: A Tutorial.” 1604.00772.

Dictionary

This Tuner can be instantiated with the associated sugar function tnr():

tnr("cmaes")

Progress Bars

$optimize() supports progress bars via the package progressr combined with a bbotk::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").

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::OptimizerBatchCmaes which can be applied on any black box optimization problem. See also the documentation of bbotk.

Parameters

start_values

character(1)
Create "random" start values or based on "center" of search space? In the latter case, it is the center of the parameters before a trafo is applied. If set to "custom", the start values can be passed via the start parameter.

start

numeric()
Custom start values. Only applicable if start_values parameter is set to "custom".

seed

integer(1)
Seed of the random number generator of libcmaes. Unset by default, in which case the generator is seeded from R and the optimization is reproducible with set.seed().

All remaining parameters are passed to libcmaesr::cmaes_control(), see there for their meaning. Note that we have removed all control parameters which refer to the termination of the algorithm and where our terminators allow to obtain the same behavior, i.e. max_fevals, max_iter, and ftarget. The internal convergence criteria of the algorithm still apply, so the optimization can stop before the Terminator is triggered.

Batch evaluation

The optimizer evaluates a whole generation of lambda points in one batch. The Terminator is only checked between generations, so the number of evaluations can exceed the budget of TerminatorEvals by up to lambda - 1 points.

Resources

There are several sections about hyperparameter optimization in the mlr3book.

The gallery features a collection of case studies and demos about optimization.

The cheatsheet summarizes the most important functions of mlr3tuning.

See also

Other Tuner: Tuner, mlr_tuners, mlr_tuners_design_points, mlr_tuners_gensa, mlr_tuners_grid_search, mlr_tuners_internal, mlr_tuners_irace, mlr_tuners_nloptr, mlr_tuners_random_search

Super classes

Tuner -> TunerBatch -> TunerBatchFromOptimizerBatch -> TunerBatchCmaes

Methods

Public methods

Inherited methods

TunerBatchCmaes$new()

Creates a new instance of this R6 class.

Usage

TunerBatchCmaes$new()

TunerBatchCmaes$clone()

The objects of this class are cloneable with this method.

Usage

TunerBatchCmaes$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# example only runs if libcmaesr is available
if (mlr3misc::require_namespaces("libcmaesr", quietly = TRUE)) {
# Hyperparameter Optimization

# load learner and set search space
learner = lrn("classif.rpart",
  cp = to_tune(1e-04, 1e-1, logscale = TRUE),
  minsplit = to_tune(p_dbl(2, 128, trafo = as.integer)),
  minbucket = to_tune(p_dbl(1, 64, trafo = as.integer))
)

# run hyperparameter tuning on the Palmer Penguins data set
instance = tune(
  tuner = tnr("cmaes"),
  task = tsk("penguins"),
  learner = learner,
  resampling = rsmp("holdout"),
  measure = msr("classif.ce"),
  term_evals = 10)

# best performing hyperparameter configuration
instance$result

# all evaluated hyperparameter configuration
as.data.table(instance$archive)

# fit final model on complete data set
learner$param_set$values = instance$result_learner_param_vals
learner$train(tsk("penguins"))
}