ActiveLearningSelection<T>

Models & Types · AiDotNet.ActiveLearning

The result of an active-learning pass: a full ranking of the unlabeled pool plus the diversity-aware batch chosen for labeling.

For Beginners

This tells you which unlabeled samples to label next, and why: each sample’s uncertainty, whether it made the batch, and the order it was chosen.

How It Works

Beyond a plain uncertainty ranking, the batch is chosen with a diversity/redundancy penalty (BADGE / facility-location style) so it covers the pool rather than collecting many near-identical uncertain samples — the failure mode of naive top-N uncertainty sampling.

Constructors

Constructor Summary
ActiveLearningSelection()

Properties

Property Summary
IReadOnlyList<ActiveLearningCandidate<T>> Ranking { get; set; } Gets the whole pool ranked, most-informative first.
double DiversityWeight { get; set; } Gets the diversity weight used to penalize redundancy (0 = pure uncertainty, higher = more diversity pressure).
int BatchSize { get; set; } Gets the requested batch size.
int[] SelectedIndices { get; set; } Gets the selected sample indices (into the pool), in selection order.
string RepresentationSpace { get; set; } Gets the space in which diversity was measured: "ModelRepresentation" when a model embedding was available, otherwise "InputFeatures".
string StrategyName { get; set; } Gets the configured strategy’s name.