FMeasure<T>

Models & Types · AiDotNet.Clustering.Evaluation

F-Measure (F-Score) for comparing clustering results against ground truth.

Implements: IExternalClusterMetric<T>

For Beginners

F-Measure asks “How well do clusters match classes?”

For each true class:

  • Find which cluster matches it best
  • Calculate how well that cluster captures the class

Precision: “Of the cluster, how many belong to this class?” Recall: “Of this class, how many are in this cluster?” F-Measure: Balances both (harmonic mean)

Higher is better! 1.0 = perfect, 0 = no agreement.

How It Works

The F-Measure combines precision and recall for clustering evaluation. For each true class, it finds the best matching cluster, then averages the F-scores weighted by class sizes.

F(C_i, K_j) = (2 * Precision * Recall) / (Precision + Recall) Where:

  • Precision = |C_i ∩ K_j| / |K_j|
  • Recall = |C_i ∩ K_j| / |C_i|

Constructors

Constructor Summary
FMeasure(double beta) Initializes a new FMeasure instance.

Methods

Method Summary
Dictionary<ValueTuple<int, int>, double> ComputeMatrix(Vector<T> trueLabels, Vector<T> predictedLabels) Computes the F-Measure matrix between all classes and clusters.
double Compute(Vector<T> trueLabels, Vector<T> predictedLabels)
double ComputeBCubed(Vector<T> trueLabels, Vector<T> predictedLabels) Computes pair-counting based F-Measure (BCubed F-Measure).