Round Metrics

The RoundMetrics dataclass captures evaluation metrics computed at the end of each active learning round, including model performance scores and acquisition statistics. These per-round snapshots are collected into a Results object at experiment completion.

class alf_core.dataclasses.round_metrics.RoundMetrics(round, metrics=<factory>, training_history=<factory>)

Bases: object

All metrics for a single acquisition round.

round is the canonical round number — it is not duplicated inside metrics. training_history carries per-epoch SurrogateEpochMetrics objects for backends that support step-based logging

round

The round number this instance describes.

metrics

Flat dict of scalar metrics for this round (e.g. tell_time, surrogate/test_spearman, dataset/num_train).

training_history

Per-epoch metrics recorded during surrogate training in this round. Empty list when no training occurred (e.g. zero-shot tasks) or before Surrogate.fit() has been called.

metrics: dict[str, Any]

round: int

training_history: list[SurrogateEpochMetrics]