Skip to contents

Search Metrics

Source: R/search_metrics.R
search_metrics.Rd

Built-in metric functions for tracking search progress. These functions are used by MetricsTracker to compute per-batch summaries.

Details

All metric functions follow the signature:

function(archive, surrogate = NULL, ...)

They return a single numeric value. The ... can receive additional context like test data, known optimum, evaluation grid, etc.

Optimization Metrics

  • metric_best_y: Best observed value (for minimization)

  • metric_worst_y: Worst observed value

  • metric_regret: Gap between current best and known optimum

  • metric_simple_regret: Same as regret (alias)

Learning Metrics

  • metric_model_rmse: Surrogate RMSE on test data

  • metric_model_mae: Surrogate MAE on test data

  • metric_model_r2: Surrogate R-squared on test data

  • metric_mean_variance: Average prediction variance on a grid

  • metric_max_variance: Maximum prediction variance on a grid

  • metric_integrated_variance: Integral of variance over domain

Progress Metrics

  • metric_n_evals: Total evaluations (built-in to tracker)

  • metric_improvement_rate: Improvement per evaluation