Implementation:Scikit learn Scikit learn BisectingKMeans

Knowledge Sources	Scikit_learn Scikit-learn Docs
Domains	Clustering, Divisive Clustering
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for performing bisecting K-Means clustering provided by scikit-learn.

Description

BisectingKMeans is a variant of KMeans that uses a divisive (top-down) hierarchical approach. It starts with all data points in a single cluster and iteratively bisects the cluster with the largest score (based on either inertia or number of data points) using standard 2-means until the desired number of clusters is reached. Internally it builds a binary tree structure (_BisectingTree) to track the hierarchical cluster decomposition. It inherits from _BaseKMeans and supports both Lloyd and Elkan algorithms.

Usage

Use BisectingKMeans when you want a hierarchical variant of K-Means that produces more regular clusters with a consistent structure. It is well-suited when you need a divisive clustering approach, or when standard K-Means initialization strategies yield poor results due to data geometry.

Code Reference

Source Location

• Repository: scikit-learn
• File: sklearn/cluster/_bisect_k_means.py

Signature

class BisectingKMeans(_BaseKMeans):
    def __init__(
        self,
        n_clusters=8,
        *,
        init="random",
        n_init=1,
        random_state=None,
        max_iter=300,
        verbose=0,
        tol=1e-4,
        copy_x=True,
        algorithm="lloyd",
        bisecting_strategy="biggest_inertia",
    ):

Import

from sklearn.cluster import BisectingKMeans

I/O Contract

Inputs

Name	Type	Required	Description
n_clusters	int	No	Number of clusters to form. Default is 8.
init	str or callable	No	Initialization method: "k-means++", "random", or a callable. Default is "random".
n_init	int	No	Number of inner k-means runs per bisection. Default is 1.
random_state	int or RandomState	No	Random state for reproducibility. Default is None.
max_iter	int	No	Maximum iterations per inner k-means run. Default is 300.
verbose	int	No	Verbosity level. Default is 0.
tol	float	No	Relative tolerance for convergence. Default is 1e-4.
copy_x	bool	No	Whether to copy input data. Default is True.
algorithm	str	No	K-Means algorithm to use: "lloyd" or "elkan". Default is "lloyd".
bisecting_strategy	str	No	Strategy for choosing cluster to bisect: "biggest_inertia" or "largest_cluster". Default is "biggest_inertia".

Outputs

Name	Type	Description
cluster_centers_	ndarray of shape (n_clusters, n_features)	Coordinates of cluster centers.
labels_	ndarray of shape (n_samples,)	Cluster labels for each sample.
inertia_	float	Sum of squared distances of samples to their closest cluster center.
n_iter_	int	Number of bisection iterations performed.

Usage Examples

Basic Usage

from sklearn.cluster import BisectingKMeans
import numpy as np

X = np.array([[1, 2], [1, 4], [1, 0],
              [10, 2], [10, 4], [10, 0]])

bisect_km = BisectingKMeans(n_clusters=2, random_state=0).fit(X)
print(bisect_km.labels_)
print(bisect_km.cluster_centers_)

Related Pages

• Principle:Scikit_learn_Scikit_learn_Clustering