MarginSampling<T>

Models & Types · AiDotNet.ActiveLearning

Implements margin sampling for active learning sample selection.

Implements: IActiveLearningStrategy<T>

For Beginners

Margin sampling selects samples where the difference between the top two predicted class probabilities is smallest. A small margin indicates the model is uncertain between two classes, suggesting the sample is near the decision boundary.

How It Works

Formula: margin = P(y₁|x) - P(y₂|x), where y₁ and y₂ are the most likely classes.

Selection: Select samples with smallest margins (transformed as 1 - margin for consistency with other strategies where higher = more informative).

Advantages:

Complexity: O(n × c) where n=pool size, c=number of classes.

Reference: Settles, B. (2012). “Active Learning.” Morgan & Claypool Publishers.

Constructors

Constructor Summary
MarginSampling() Initializes a new instance of the MarginSampling class.

Properties

Property Summary
bool UseBatchDiversity { get; set; }
string Name { get; }

Methods

Method Summary
Dictionary<string, T> GetSelectionStatistics()
Vector<T> ComputeInformativenessScores(IFullModel<T, Tensor<T>, Tensor<T>> model, Tensor<T> unlabeledPool)
int[] SelectSamples(IFullModel<T, Tensor<T>, Tensor<T>> model, Tensor<T> unlabeledPool, int batchSize)