MMPCSelector<T>

Models & Types ยท AiDotNet.Preprocessing.FeatureSelection.Causal

MMPC (Max-Min Parents and Children) based Feature Selection.

Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> โ†’ MMPCSelector<T>

Implements: IFixedCountFeatureSelector<T>, ISupervisedFeatureSelector<T>

For Beginners

MMPC finds features that are direct causes or effects of the target (parents/children in a causal graph). It uses conditional independence tests to identify these relationships. The result is a minimal set of truly relevant features, removing spurious correlations.

How It Works

Selects features using the MMPC algorithm, which identifies the Markov blanket of the target - the minimal set of features needed for prediction.

Note: The alpha parameter in this implementation is used as a minimum association threshold (correlation magnitude), not as a p-value significance level as in traditional MMPC. Features with association below this threshold are considered conditionally independent.

Constructors

Constructor Summary
MMPCSelector(int nFeaturesToSelect, double minAssociationThreshold, int[] columnIndices)

Properties

Property Summary
Vector<T> AssociationScores { get; }
bool SupportsInverseTransform { get; }
double MinAssociationThreshold { get; }
int NFeaturesToSelect { get; }
int[] SelectedIndices { get; }

Methods

Method Summary
Matrix<T> FitTransform(Matrix<T> data, Vector<T> target)
bool[] GetSupportMask()
string[] GetFeatureNamesOut(string[] inputFeatureNames)
void Fit(Matrix<T> data, Vector<T> target)