Build an LCE Task from an Archive and a Performance Table

Builds a TaskLCE by joining an optimization archive with a per-batch performance table: one task row per archive evaluation, with the selected performance column as target. This is the assembly used by CallbackSurrogatePerformance$task() and replay_surrogate_performance(); it is exported for benchmark harnesses that record archives and curves separately and rebuild tasks later.

Usage

task_lce_from_perf(
  archive,
  perf,
  measure,
  measure_obj = NULL,
  pool = NULL,
  link = "identity",
  id,
  label = NA_character_
)

Arguments

archive: (bbotk::Archive)
Archive of the originating run; supplies the archive_x / archive_y columns, batch_nr, and the search space / codomain provenance.
perf: (data.table::data.table)
Per-batch performance table with a batch_nr column plus one column per measure. Archive batches without a matching perf row are dropped.
measure: (character(1))
Name of the perf column to use as task target.
measure_obj: (mlr3::Measure | NULL)
The regression measure that produced the target column; carried into the task so replay extrapolators can score on the same scale.
pool: (data.table::data.table | NULL)
Candidate pool of the originating run (search-space columns only), for pool-restricted replay.
link: (character(1))
Name of the predictive lce_link for the resulting task.
id: (character(1))
Task id.
label: (character(1))
Optional task label.

Value

TaskLCE.