SelectFromModel<T>

Models & Types · AiDotNet.Preprocessing.FeatureSelection.Embedded

Selects features based on importance weights from an external model or scorer.

Inheritance: TransformerBase<T, Matrix<T>, Matrix<T>> → SelectFromModel<T>

Implements: ISupervisedFeatureSelector<T>

For Beginners

This works with any model that produces feature importances:

  • Random forests give importance based on how much each feature reduces error
  • Linear models give coefficients showing feature influence
  • Features below the threshold are removed

How It Works

SelectFromModel selects features based on importance scores, typically from a fitted model. Features with importance above a threshold are kept.

The threshold can be specified as:

  • An absolute value
  • “mean” - the mean of feature importances
  • “median” - the median of feature importances

Constructors

Constructor Summary
SelectFromModel(Func<Matrix<T>, Vector<T>, Vector<T>> importanceFunc, SelectFromModelThreshold thresholdType, double thresholdValue, int? maxFeatures, int[] columnIndices) Creates a new instance with a function to compute feature importances.

Properties

Property Summary
T Threshold { get; } Gets the computed threshold value.
Vector<T> FeatureImportances { get; } Gets the feature importances used for selection.
bool SupportsInverseTransform { get; } Gets whether this transformer supports inverse transformation.
int[] SelectedIndices { get; } Gets the indices of selected features.

Methods

Method Summary
Matrix<T> FitTransform(Matrix<T> data, Vector<T> target) Fits and transforms the data.
SelectFromModel<T> FromImportances(Vector<T> featureImportances, SelectFromModelThreshold thresholdType, double thresholdValue, int? maxFeatures, int[] columnIndices) Creates a new instance with precomputed feature importances.
bool[] GetSupportMask() Gets the support mask indicating which features are selected.
string[] GetFeatureNamesOut(string[] inputFeatureNames) Gets the output feature names after transformation.
void Fit(Matrix<T> data, Vector<T> target) Fits the selector by computing feature importances.