Skip to contents

Integrated Variance Metric

Source: R/search_metrics.R
metric_integrated_variance.Rd

Approximates the integral of prediction variance over the domain. This is the continuous version of mean variance, accounting for domain volume.

Usage

metric_integrated_variance(
  archive,
  surrogate = NULL,
  grid = NULL,
  search_space = NULL,
  ...
)

Arguments

archive

(bbotk::ArchiveBatch) The archive (may be used for normalization).

surrogate

(mlr3mbo::Surrogate) The fitted surrogate model.

grid

(data.table) Points at which to evaluate variance. If NULL, generates a default grid from the search space.

search_space

(paradox::ParamSet) Used to generate grid if not provided.

...

Ignored.

Value

numeric(1) The integrated variance.

Details

For a uniform grid, this equals mean_variance * domain_volume. More sophisticated integration (e.g., Monte Carlo) may be added.