sklekmeans 0.1.1

Equilibrium K-Means (EKMeans) clustering algorithms compatible with scikit-learn

sklekmeans - Equilibrium K-Means for scikit-learn

sklekmeans provides batch and mini-batch implementations of the Equilibrium K-Means (EKMeans) clustering algorithm. The method introduces an equilibrium weighting scheme that can yield improved robustness on imbalanced datasets compared to standard k-means.

Features

• Drop-in scikit-learn compatible estimators: EKMeans, MiniBatchEKMeans.
• Supports Euclidean and Manhattan distances.
• Heuristic alpha selection via alpha='dvariance'.
• Mini-batch variant with accumulation or online update modes.
• Soft memberships (membership) and equilibrium weights (W_).

Installation

The package is available on PyPI. Install the base package:

pip install sklekmeans

Optional extras:

• With numba acceleration (recommended for speed):

pip install "sklekmeans[speed]"

From source (latest main):

• Basic installation

git clone https://github.com/ydcnanhe/sklearn-ekmeans.git
cd sklearn-ekmeans
pip install .

• Or in editable mode

pip install -e .

• With numba acceleration

pip install -e .[speed]

• Development tools (tests, lint):

pip install -e .[dev]

• Docs build dependencies:

pip install -e .[docs]

• Everything (dev + docs + speed):

pip install -e .[all]

Quick Start

from sklekmeans import EKMeans
import numpy as np

X = np.random.rand(200, 2)
ekm = EKMeans(n_clusters=3, random_state=0, alpha='dvariance').fit(X)
print(ekm.cluster_centers_)

Mini-batch variant with multiple initializations and selection of the best run:

from sklekmeans import MiniBatchEKMeans
mb = MiniBatchEKMeans(n_clusters=3, batch_size=256, max_epochs=20, n_init=5, random_state=0)
mb.fit(X)
print(mb.cluster_centers_)

Documentation

The latest HTML documentation is hosted on GitHub Pages:

ydcnanhe.github.io/sklearn-ekmeans

Badges above reflect build status; if the link 404s, wait for the docs CI to finish.

PyPI project page: https://pypi.org/project/sklekmeans/

Build and publish (maintainers)

Local build of artifacts:

python -m pip install --upgrade build twine
python -m build
python -m twine check dist/*

Publishing to PyPI is automated via GitHub Actions (Trusted Publishing). See PUBLISHING.md.

References

• [1] Y. He. An Equilibrium Approach to Clustering: Surpassing Fuzzy C-Means on Imbalanced Data, IEEE Transactions on Fuzzy Systems, 2025.
• [2] Y. He. Imbalanced Data Clustering Using Equilibrium K-Means, arXiv, 2024.

License

BSD 3-Clause