Bootstrap632Splitter<T>

Models & Types · AiDotNet.Preprocessing.DataPreparation.Splitting.Bootstrap

.632 Bootstrap that provides bias-corrected error estimation.

Inheritance: DataSplitterBase<T> → BootstrapSplitter<T> → Bootstrap632Splitter<T>

For Beginners

Regular bootstrap error estimates can be biased (too optimistic). The .632 bootstrap corrects this by combining training error and OOB error.

How It Works

The Formula: Error_632 = 0.368 × TrainingError + 0.632 × OOB_Error

Why .632? This number comes from the probability that a sample is NOT in a bootstrap sample: (1 - 1/n)^n ≈ 1/e ≈ 0.368 So approximately 63.2% of unique samples end up in training.

Note: This splitter generates the same splits as regular bootstrap. The .632 error calculation should be done during model evaluation.

Constructors

Constructor Summary
Bootstrap632Splitter(int nIterations, int randomSeed) Creates a new .632 bootstrap splitter.

Properties

Property Summary
string Description { get; }