K-means clustering implementation whereby a minimum and/or maximum size for each cluster can be specified.
This K-means implementation modifies the cluster assignment step (E in EM)
by formulating it as a Minimum Cost Flow (MCF) linear network
optimisation problem. This is then solved using a cost-scaling
push-relabel algorithm and uses Google's Operations Research tools's
SimpleMinCostFlow
which is a fast C++ implementation.
This package is inspired by Bradley et al.. The original Minimum Cost Flow (MCF) network proposed by Bradley et al. has been modified so maximum cluster sizes can also be specified along with minimum cluster size.
The code is based on scikit-lean's KMeans
and implements the same API with modifications.
Ref:
You can install the k-means-constrained from PyPI:
pip install k-means-constrained
It is supported on Python 3.10, 3.11 and 3.12. Previous versions of k-means-constrained support older versions of Python and Numpy.
More details can be found in the API documentation.
>>> from k_means_constrained import KMeansConstrained
>>> import numpy as np
>>> X = np.array([[1, 2], [1, 4], [1, 0],
... [4, 2], [4, 4], [4, 0]])
>>> clf = KMeansConstrained(
... n_clusters=2,
... size_min=2,
... size_max=5,
... random_state=0
... )
>>> clf.fit_predict(X)
array([0, 0, 0, 1, 1, 1], dtype=int32)
>>> clf.cluster_centers_
array([[ 1., 2.],
[ 4., 2.]])
>>> clf.labels_
array([0, 0, 0, 1, 1, 1], dtype=int32)
k-means-constrained is a more complex algorithm than vanilla k-means and therefore will take longer to execute and has worse scaling characteristics.
Given a number of data points $n$ and clusters $c$, the time complexity of:
This assumes a constant number of algorithm iterations and data-point features/dimensions.
If you consider the case where $n$ is the same order as $c$ ($n \backsim c$) then:
Below is a runtime comparison between k-means and k-means-constrained whereby the number of iterations, initializations, multi-process pool size and dimension size are fixed. The number of clusters is also always one-tenth the number of data points $n=10c$. It is shown above that the runtime is independent of the minimum or maximum cluster size, and so none is included below.
1: Ortools states the time complexity of their cost-scaling push-relabel algorithm for the min-cost flow problem as $\mathcal{O}(n^2m\log(nC))$ where $n$ is the number of nodes, $m$ is the number of edges and $C$ is the maximum absolute edge cost.
If you use this software in your research, please use the following citation:
@software{Levy-Kramer_k-means-constrained_2018,
author = {Levy-Kramer, Josh},
month = apr,
title = {{k-means-constrained}},
url = {https://github.com/joshlk/k-means-constrained},
year = {2018}
}