Hierarchical clustering in ML Python

Introduction

Hierarchical clustering groups similar data points step-by-step to show how they relate in a tree-like structure.

When you want to see how data naturally groups without deciding the number of groups first.

When you need a visual tree (dendrogram) to understand data relationships.

When working with small to medium datasets to explore data structure.

When you want to cluster data but don't know how many clusters to expect.

When you want to combine or split clusters at different levels of detail.

Syntax

ML Python

from sklearn.cluster import AgglomerativeClustering

model = AgglomerativeClustering(n_clusters=number, linkage='ward')
model.fit(data)
labels = model.labels_

n_clusters sets how many groups you want at the end.

linkage defines how to measure distance between clusters; common options are 'ward', 'complete', and 'average'.

Examples

Groups data into 3 clusters using Ward's method, which minimizes variance inside clusters.

ML Python

model = AgglomerativeClustering(n_clusters=3, linkage='ward')

Groups data into 2 clusters using complete linkage, which considers the farthest points between clusters.

ML Python

model = AgglomerativeClustering(n_clusters=2, linkage='complete')

Stops clustering when clusters are farther than 1.5 units apart, letting the algorithm decide the number of clusters.

ML Python

model = AgglomerativeClustering(n_clusters=None, distance_threshold=1.5)

Sample Program

This program creates simple data with 3 groups, applies hierarchical clustering to find 3 clusters, and prints the cluster each point belongs to.

ML Python

from sklearn.cluster import AgglomerativeClustering
from sklearn.datasets import make_blobs

# Create sample data with 3 centers
X, _ = make_blobs(n_samples=15, centers=3, random_state=42)

# Create and fit hierarchical clustering model
model = AgglomerativeClustering(n_clusters=3, linkage='ward')
model.fit(X)

# Print cluster labels for each point
print('Cluster labels:', model.labels_)

Important Notes

Hierarchical clustering can be slow on very large datasets because it compares all points step-by-step.

The dendrogram is a helpful visual to decide the number of clusters by cutting the tree at different heights.

Choosing the right linkage method affects how clusters form; try different ones to see what fits your data best.

Summary

Hierarchical clustering builds a tree of clusters by joining or splitting data points stepwise.

It does not require you to pick the number of clusters upfront if you use distance thresholds.

Useful for exploring data structure and visualizing relationships with dendrograms.