Online Robotics Classes for Kids in Athens

Chapter 136 – Hierarchical Clustering with Scikit-Learn (Part 1)

1. Agglomerative Flow

Agglomerative clustering is unsupervised and bottom-up: every sample starts alone, closest pairs merge until n_clusters is reached. No target labels are used—only distances in the feature matrix.

2. Core Hyperparameters

Parameter	Role
n_clusters	Final number of groups (e.g. 3)
metric / affinity	Distance (euclidean, manhattan, …)
linkage	ward (min variance), complete (max pair), average (mean pair)

3. Iris Dataset

150 rows × 4 features (sepal/petal length & width in cm). Same units across columns → no StandardScaler required before distance-based clustering.

4. Implementation Pipeline

import pandas as pd
from sklearn.datasets import load_iris
from sklearn.cluster import AgglomerativeClustering

iris = load_iris()
X = pd.DataFrame(iris.data, columns=iris.feature_names)
print(X.shape)  # (150, 4)

hc = AgglomerativeClustering(n_clusters=3, linkage="ward")
labels = hc.fit_predict(X)
X["cluster"] = labels
print(X["cluster"].value_counts())

5. Typical Cluster Counts (Ward, k=3)

• Cluster 0: 64 samples
• Cluster 1: 50 samples
• Cluster 2: 36 samples

IDs are arbitrary (0/1/2). Domain experts can later map them to species names.

6. Pipeline Summary

Step	Tool	Outcome
Load	load_iris()	(150, 4) matrix
Frame	pd.DataFrame	Named columns
Model	AgglomerativeClustering(n_clusters=3, linkage="ward")	Configured HC
Fit	.fit_predict(X)	150 integer labels
Inspect	.value_counts()	64 / 50 / 36