FairFeatureSelector<T>
Models & Types · AiDotNet.Preprocessing.FeatureSelection.Fairness
Fair Feature Selection that balances predictive power with fairness constraints.
Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> → FairFeatureSelector<T>
Implements: IFixedCountFeatureSelector<T>, ISupervisedFeatureSelector<T>
For Beginners
In ML fairness, we want models that don’t discriminate based on protected attributes like race or gender. This selector chooses features that are good for prediction but have low correlation with a specified protected attribute. This helps build fairer models by removing proxies for protected groups.
How It Works
Selects features that maximize predictive accuracy while minimizing correlation with protected attributes to promote fair predictions.
Constructors
| Constructor | Summary |
|---|---|
FairFeatureSelector(int nFeaturesToSelect, int protectedAttributeIndex, double fairnessWeight, int[] columnIndices) |
Properties
| Property | Summary |
|---|---|
Vector<T> CombinedScores { get; } |
|
Vector<T> FairnessScores { get; } |
|
Vector<T> PredictiveScores { get; } |
|
bool SupportsInverseTransform { get; } |
|
double FairnessWeight { get; } |
|
int NFeaturesToSelect { get; } |
|
int ProtectedAttributeIndex { get; } |
|
int[] SelectedIndices { get; } |
Methods
| Method | Summary |
|---|---|
Matrix<T> FitTransform(Matrix<T> data, Vector<T> target) |
|
T GetDisparateImpact(Matrix<T> data, int featureIndex) |
Computes the disparate impact ratio for a specific feature. |
bool[] GetSupportMask() |
|
string[] GetFeatureNamesOut(string[] inputFeatureNames) |
|
void Fit(Matrix<T> data, Vector<T> target) |