OSFS<T>

Models & Types · AiDotNet.Preprocessing.FeatureSelection.Online

Online Streaming Feature Selection (OSFS) of Wu et al. 2013 (“Online Feature Selection with Streaming Features”, IEEE TPAMI 35(5)).

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

Implements: IFixedCountFeatureSelector<T>, ISupervisedFeatureSelector<T>

For Beginners

Imagine features arriving one by one like a stream. For each new feature, OSFS asks: “Is this related to the target at all?” If yes, it keeps the feature and then re-checks every kept feature: one that tells nothing more about the target once some of the others are known is dropped.

How It Works

The features arrive one at a time, in column order. Each new feature X goes through the paper’s two steps:

  • Online relevance analysis: X is discarded when it is independent of the target C,

Ind(C, X | {}).

  • Online redundancy analysis: otherwise X joins the best candidate features BCF, and every Y in BCF

is removed when some subset S of the other BCF features (|S| up to MaxConditioningSetSize) makes it conditionally independent of the target, Ind(C, Y | S).

The independence tests are Fisher’s z test on the partial correlation at level Alpha, as in the authors’ LOFS toolbox (osfs_z): W = sqrt(n - |S| - 3) atanh(r), independent when the two-sided p-value 2 Phi(-|W|) exceeds alpha.

OSFS decides the subset size itself. NFeaturesToSelect caps it: when BCF ends larger, the features with the strongest correlation with the target are kept.

Constructors

Constructor Summary
OSFS(int nFeaturesToSelect, double alpha, int maxConditioningSetSize, int[] columnIndices) Creates an OSFS selector.

Properties

Property Summary
Vector<T> FeatureScores { get; } The absolute Pearson correlation of every feature with the target.
bool SupportsInverseTransform { get; }
double Alpha { get; } The significance level of the conditional-independence tests.
int MaxConditioningSetSize { get; } The largest conditioning set the redundancy analysis tries.
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)