IBatchStrategy<T, TInput, TOutput>

Interfaces · AiDotNet.ActiveLearning.Interfaces

Interface for batch selection strategies in active learning.

Known implementations: IClusteringBatchStrategy<T, TInput, TOutput>, IGradientBatchStrategy<T, TInput, TOutput>, ISubmodularBatchStrategy<T, TInput, TOutput>, RankedBatchStrategy<T, TInput, TOutput>

For Beginners

When selecting multiple samples for labeling at once (batch mode), we need to ensure diversity among the selected samples. Simply taking the top-N by score might select very similar samples, which wastes labeling budget.

How It Works

Batch Selection Approaches:

Properties

Property Summary
T DiversityTradeoff { get; set; } Gets or sets the trade-off parameter between informativeness and diversity.
string Name { get; } Gets the name of the batch selection strategy.

Methods

Method Summary
T ComputeDiversity(TInput sample1, TInput sample2) Computes pairwise diversity between samples.
int[] SelectBatch(int[] candidateIndices, Vector<T> scores, IDataset<T, TInput, TOutput> unlabeledPool, int batchSize) Selects a diverse batch of samples from candidates.