AdversarialValidationSplitter<T>
Models & Types · AiDotNet.Preprocessing.DataPreparation.Splitting.Specialized
Adversarial validation splitter that identifies samples most similar to test distribution.
Inheritance: DataSplitterBase<T> → AdversarialValidationSplitter<T>
For Beginners
Sometimes your training and test data come from different distributions (e.g., train from 2022, test from 2023). Adversarial validation helps detect this problem.
How It Works
How It Works:
- Train a classifier to distinguish train vs test samples
- If the classifier performs well (AUC > 0.5), there’s distribution shift
- Use the classifier’s predictions to create a more realistic validation set
- Put samples most “test-like” into validation
Practical Use: This splitter doesn’t actually train a classifier - it creates indices that you can use after running adversarial validation externally. You provide the probability scores from your adversarial classifier.
When to Use:
- Kaggle competitions where public/private test differs from train
- Time-based production scenarios
- Any situation with potential train/test distribution shift
Constructors
| Constructor | Summary |
|---|---|
AdversarialValidationSplitter(double[] testProbabilities, double testSize, int randomSeed) |
Creates an adversarial validation splitter. |
Properties
| Property | Summary |
|---|---|
string Description { get; } |
Methods
| Method | Summary |
|---|---|
DataSplitResult<T> Split(Matrix<T> X, Vector<T> y) |