SimilarityBasedSplitter<T>

Models & Types ยท AiDotNet.Preprocessing.DataPreparation.Splitting.Specialized

Similarity-based splitter that splits data based on sample similarity scores.

Inheritance: DataSplitterBase<T> โ†’ SimilarityBasedSplitter<T>

For Beginners

This splitter uses similarity between samples to create splits. Samples that are very similar to training data are placed in the test set to evaluate interpolation, while dissimilar samples test extrapolation capability.

How It Works

Modes:

  • Interpolation Test: Test samples similar to training (tests within-distribution generalization)
  • Extrapolation Test: Test samples dissimilar to training (tests out-of-distribution performance)

When to Use:

  • When you want to specifically test interpolation vs extrapolation
  • For robustness evaluation
  • When similarity structure in data is meaningful

Constructors

Constructor Summary
SimilarityBasedSplitter(double testSize, double threshold, bool testSimilar, bool shuffle, int randomSeed) Creates a new similarity-based splitter.

Properties

Property Summary
string Description { get; }

Methods

Method Summary
DataSplitResult<T> Split(Matrix<T> X, Vector<T> y)