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; } |