Abstract base class for performance measures evaluating the surrogate-quality forecast produced by a LearnerLCE. Concrete measures compare the predicted per-batch performance to the recorded performance.
Measures defined via this base class always aggregate predictions and truths to one value per batch (the within-batch mean) before computing the per-batch loss. Multiple archive rows belong to the same batch and share the same surrogate-quality value, so the aggregation collapses them without affecting the loss; if a future learner predicts differently within a batch, the mean is the natural reduction.
Creates a new LCE measure.
Arguments
- id
(
character(1))
Measure id.- param_set
- range
(
numeric(2))
Theoretical range of values.- minimize
(
logical(1)).- average
(
character(1)).- aggregator
(
function()orNULL).- properties
(
character()).- predict_type
(
character(1))
Any lce predict type (see PredictionLCE), e.g."response","se","quantiles".- predict_sets
(
character()).- task_properties
(
character()).- packages
(
character()).- label
(
character(1)).- man
(
character(1)).