ActiveLearnerConfig<T>

Options & Configuration · AiDotNet.ActiveLearning.Config

Comprehensive configuration for active learning. All properties are nullable - null values use industry-standard defaults.

For Beginners

Active learning is a machine learning paradigm where the algorithm actively selects which data points should be labeled by an oracle (human expert). This is particularly useful when labeling data is expensive or time-consuming.

How It Works

How Active Learning Works:

Key Concepts:

Constructors

Constructor Summary
ActiveLearnerConfig()

Properties

Property Summary
ClusteringMethod? DiversityClustering { get; set; } Clustering method for diversity-based sampling.
ColdStartStrategy? ColdStart { get; set; } Strategy for selecting initial labeled samples.
DisagreementMeasure? DisagreementMeasure { get; set; } How to measure disagreement in the committee.
DistanceMetric? CoresetDistance { get; set; } Distance metric for coreset selection.
GradientApproximation? GradientMethod { get; set; } Gradient approximation method for expected model change.
QueryStrategyType? QueryStrategy { get; set; } Primary query strategy to use for sample selection.
StoppingCriterionType? StoppingCriterion { get; set; } Type of stopping criterion to use.
T DiversityWeight { get; set; } Weight for diversity in hybrid uncertainty-diversity strategies.
T LearningRate { get; set; } Learning rate for model training.
T McDropoutRate { get; set; } Dropout rate for Monte Carlo Dropout.
T MinAccuracyGain { get; set; } Minimum accuracy gain required to continue.
T TestSetFraction { get; set; } Fraction of data to use as held-out test set.
UncertaintyMeasure? UncertaintyMeasure { get; set; } Uncertainty measure for uncertainty-based sampling.
bool? BatchBaldGreedy { get; set; } Whether to use greedy approximation for BatchBALD.
bool? CoresetGreedy { get; set; } Whether to use greedy k-center algorithm.
bool? EnableAutoStop { get; set; } Whether to enable automatic stopping before budget exhaustion.
bool? EvaluatePerIteration { get; set; } Whether to evaluate on a held-out test set after each iteration.
bool? ExpectedErrorReduction { get; set; } Use expected error reduction for query selection.
bool? HandleLabelNoise { get; set; } Whether to handle label noise in the oracle’s responses.
bool? StratifiedInitial { get; set; } Whether to use stratified sampling for initial selection.
bool? WarmStart { get; set; } Enable warm starting between iterations (keep model weights).
int? BatchBaldCandidates { get; set; } Number of candidates to consider for BatchBALD selection.
int? CommitteeSize { get; set; } Number of models in the committee for QBC.
int? EpochsPerIteration { get; set; } Number of training epochs per active learning iteration.
int? InitialPoolSize { get; set; } Number of initially labeled samples to start with.
int? MaxBudget { get; set; } Maximum total samples that can be labeled (labeling budget).
int? McDropoutSamples { get; set; } Number of Monte Carlo Dropout forward passes for BALD.
int? QueryBatchSize { get; set; } Number of samples to query in each active learning iteration.
int? Seed { get; set; } Random seed for reproducibility.
int? StoppingPatience { get; set; } Patience for stopping criteria (number of iterations without improvement).
int? TrainingBatchSize { get; set; } Batch size for training.

Methods

Method Summary
QueryStrategyType GetEffectiveQueryStrategy() Gets the effective query strategy with default fallback.
T GetEffectiveLearningRate() Gets the effective learning rate with default fallback.
int GetEffectiveInitialPoolSize() Gets the effective initial pool size with default fallback.
int GetEffectiveMaxBudget() Gets the effective max budget with default fallback.
int GetEffectiveQueryBatchSize() Gets the effective query batch size with default fallback.