intclustval

        

A lightweight Python package for internal clustering validation metrics.

intclustval provides a simple InternalClusterScore class for evaluating clustering quality using internal validation metrics.

Internal clustering validation metrics use only the input data and predicted cluster labels. They do not require ground-truth labels.

Related packages

This package is part of a small clustering-validation ecosystem:

Package	Purpose
intclustval	Internal clustering validation metrics
extclustval	External clustering validation metrics using ground-truth labels
sil-score	Exact and approximate silhouette scoring

Silhouette scores are intentionally not included in intclustval, because they are provided by the separate sil-score package.

This keeps intclustval focused on other internal validation metrics such as Calinski-Harabasz, Davies-Bouldin, inertia, Dunn Index, and Xie-Beni.

Metrics included

Internal clustering validation metrics

Attribute	Metric	Better direction
calinski_harabasz	Calinski-Harabasz score	Higher is better
davies_bouldin	Davies-Bouldin score	Lower is better
inertia	Within-cluster sum of squared distances	Lower is better for fixed number of clusters
dunn_index	Dunn Index	Higher is better
xie_beni	Xie-Beni index	Lower is better

Aliases

Attribute	Alias for
ch	calinski_harabasz
db	davies_bouldin
within_cluster_dispersion	inertia

Metadata

Attribute	Description
n_samples	Number of samples
n_features	Number of features
n_clusters	Number of clusters
labels_unique	Unique cluster labels
cluster_sizes	Number of samples in each cluster
centroids	Cluster centroids

Installation

You can install intclustval from PyPI:

pip install intclustval

Quick start

from sklearn.datasets import make_blobs
from sklearn.cluster import KMeans

from intclustval import InternalClusterScore

X, _ = make_blobs(
    n_samples=300,
    centers=3,
    cluster_std=1.0,
    random_state=42,
)

labels = KMeans(
    n_clusters=3,
    random_state=42,
    n_init=10,
).fit_predict(X)

score = InternalClusterScore(X, labels)

print(score.calinski_harabasz)
print(score.davies_bouldin)
print(score.inertia)
print(score.dunn_index)
print(score.xie_beni)

Example output:

5196.295097418395
0.21231599538998425
566.8595511244131
0.9484430301054112
0.018180444255623783

You can also access all aggregate scores as a dictionary:

scores = score.to_dict()
print(scores)

Example:

{
    "calinski_harabasz": 5196.295097418395,
    "ch": 5196.295097418395,
    "davies_bouldin": 0.21231599538998425,
    "db": 0.21231599538998425,
    "inertia": 566.8595511244131,
    "within_cluster_dispersion": 566.8595511244131,
    "dunn_index": 0.9484430301054112,
    "xie_beni": 0.018180444255623783,
}

Using silhouette scores

Silhouette scores are available in the separate sil-score package.

Install it with:

pip install sil-score

Then use:

from sil_score import (
    sil_samples,
    micro_sil_score,
    macro_sil_score,
)

sample_scores = sil_samples(X, labels)
micro_score = micro_sil_score(X, labels)
macro_score = macro_sil_score(X, labels)

print(micro_score)
print(macro_score)

The sil-score package also supports approximate silhouette scoring through its approximation argument.

Metric definitions

Calinski-Harabasz score

The Calinski-Harabasz score measures the ratio of between-cluster dispersion to within-cluster dispersion.

A higher value usually indicates better-defined clusters. It is useful for comparing different clustering solutions on the same dataset.

Davies-Bouldin score

The Davies-Bouldin score measures average similarity between each cluster and its most similar other cluster.

A lower value indicates better clustering, because it means clusters are more compact and more separated from each other.

Inertia

Inertia is the within-cluster sum of squared distances from each sample to its assigned cluster centroid.

Lower inertia means samples are closer to their cluster centers. However, inertia always decreases as the number of clusters increases, so it should mainly be used to compare solutions with different values of k on the same dataset.

Dunn Index

The Dunn Index compares the minimum distance between different clusters to the maximum diameter within any cluster.

A higher Dunn Index indicates better clustering, with clusters that are compact and well separated.

This implementation uses pairwise distances, so it may be slower for large datasets.

Xie-Beni index

The Xie-Beni index compares total within-cluster compactness to the minimum squared distance between cluster centroids.

A lower value indicates better clustering, because it means compact clusters with well-separated centers.

Notes

Internal clustering validation metrics do not use ground-truth labels. They evaluate clustering structure using only:

X
labels

For external clustering validation with ground-truth labels, use extclustval.

For silhouette-specific scoring, use sil-score.

Cached properties

InternalClusterScore uses cached properties.

This means each metric is computed once and then stored.

score = InternalClusterScore(X, labels)

score.inertia  # computed once
score.inertia  # reused from cache

If you want to evaluate different labels, create a new InternalClusterScore object:

score = InternalClusterScore(X, labels)

new_score = InternalClusterScore(X, new_labels)

Do not modify score.X or score.labels after creating the object.

Requirements

numpy
scipy
scikit-learn

License

This project is licensed under the MIT License.