Bisecting K-Means and Regular K-Means Performance Comparison#

This example shows differences between Regular K-Means algorithm and Bisecting K-Means.

While K-Means clusterings are different when increasing n_clusters, Bisecting K-Means clustering builds on top of the previous ones. As a result, it tends to create clusters that have a more regular large-scale structure. This difference can be visually observed: for all numbers of clusters, there is a dividing line cutting the overall data cloud in two for BisectingKMeans, which is not present for regular K-Means.

Bisecting K-Means : 4 clusters, Bisecting K-Means : 8 clusters, Bisecting K-Means : 16 clusters, K-Means : 4 clusters, K-Means : 8 clusters, K-Means : 16 clusters

import matplotlib.pyplot as plt

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

print(__doc__)


# Generate sample data
n_samples = 10000
random_state = 0

X, _ = make_blobs(n_samples=n_samples, centers=2, random_state=random_state)

# Number of cluster centers for KMeans and BisectingKMeans
n_clusters_list = [4, 8, 16]

# Algorithms to compare
clustering_algorithms = {
    "Bisecting K-Means": BisectingKMeans,
    "K-Means": KMeans,
}

# Make subplots for each variant
fig, axs = plt.subplots(
    len(clustering_algorithms), len(n_clusters_list), figsize=(12, 5)
)

axs = axs.T

for i, (algorithm_name, Algorithm) in enumerate(clustering_algorithms.items()):
    for j, n_clusters in enumerate(n_clusters_list):
        algo = Algorithm(n_clusters=n_clusters, random_state=random_state, n_init=3)
        algo.fit(X)
        centers = algo.cluster_centers_

        axs[j, i].scatter(X[:, 0], X[:, 1], s=10, c=algo.labels_)
        axs[j, i].scatter(centers[:, 0], centers[:, 1], c="r", s=20)

        axs[j, i].set_title(f"{algorithm_name} : {n_clusters} clusters")


# Hide x labels and tick labels for top plots and y ticks for right plots.
for ax in axs.flat:
    ax.label_outer()
    ax.set_xticks([])
    ax.set_yticks([])

plt.show()