MarginSampling¶
- class probly.evaluation.active_learning.strategies.MarginSampling[source]¶
Bases:
objectSelects unlabeled samples with the smallest margin between top-2 class probs.
A small margin indicates the model is uncertain between its top two predicted classes, making these samples most informative to label.
- select(estimator: Estimator, pool: ActiveLearningPool, n: int) ArrayLike[source]¶
Return n indices with the smallest margin in predicted probabilities.