Active Learning

The active learning module provides acquisition functions and query strategies for efficiently selecting the most informative samples to label.

Acquisition Functions

BaseAcquisition()	Base class for acquisition functions.
RandomAcquisition()	Random sampling (baseline).
EntropyAcquisition()	Entropy-based acquisition.
LeastConfidenceAcquisition()	Least confidence acquisition.
MarginAcquisition()	Margin sampling acquisition.
BALDAcquisition([num_samples])	Bayesian Active Learning by Disagreement (BALD).
VarianceRatioAcquisition([num_samples])	Variance ratio acquisition.
MeanSTDAcquisition([num_samples])	Mean standard deviation acquisition.
BatchBALDAcquisition([num_samples])	Approximate BatchBALD via individual BALD scores.

Query Strategies

UncertaintySampling(acquisition_fn[, batch_size])	Uncertainty-based sampling strategy.
DiversitySampling(acquisition_fn[, ...])	Diversity-based sampling with uncertainty.
CoreSetSelection([batch_size])	Core-Set selection for active learning.
BadgeSampling([batch_size])	BADGE (Batch Active learning by Diverse Gradient Embeddings).
QueryByCommittee(models[, batch_size, ...])	Query by Committee (QBC).

Utilities

split_labeled_unlabeled(data[, labels, ...])	Split data into labeled and unlabeled sets.
compute_diversity_penalty(selected, features)	Compute diversity penalty for selected samples.
greedy_k_center(features, k[, initial_centers])	Greedy k-center algorithm for diverse sample selection.
subsample_for_efficiency(data[, ...])	Subsample data for computational efficiency.
active_learning_loop(model, x_pool, y_pool, ...)	Run a full active learning loop.