ClusteringEntropy<T>

Models & Types · AiDotNet.Clustering.Evaluation

Entropy-based metrics for evaluating clustering against ground truth labels.

Implements: IExternalClusterMetric<T>

For Beginners

Entropy measures “How mixed are the clusters?”

Think of it like sorting socks:

  • Low entropy: Each drawer has one color (easy to find socks!)
  • High entropy: Each drawer is a random mix (chaos!)

For clustering:

  • Entropy = 0: Perfect! Each cluster has only one class
  • Entropy = log(C): Worst! Each cluster has equal parts of all classes

We want LOW entropy (pure clusters).

How It Works

Entropy measures the “disorder” or “uncertainty” in cluster assignments. Lower entropy indicates that clusters are more homogeneous with respect to the true classes.

Entropy of cluster k = -sum over all classes j of p(j|k) * log(p(j|k)) Overall entropy = sum over all clusters k of (n_k/n) * Entropy(k)

Constructors

Constructor Summary
ClusteringEntropy() Initializes a new ClusteringEntropy instance.

Methods

Method Summary
Dictionary<int, double> ComputePerCluster(Vector<T> trueLabels, Vector<T> predictedLabels) Computes entropy for each cluster.
double Compute(Vector<T> trueLabels, Vector<T> predictedLabels)
double ComputeNormalized(Vector<T> trueLabels, Vector<T> predictedLabels) Computes normalized entropy (0 = perfect, 1 = worst).