CLARANS Clustering on Interleaving Moons

This example demonstrates the behavior of k-medoids (CLARANS) on non-convex geometries like the two interleaving moons dataset.

# Authors: Ngọc Thiện Nguyễn <thiennguyen03001@gmail.com>
# License: MIT

import matplotlib.pyplot as plt
from sklearn.datasets import make_moons
from clarans import CLARANS

COLORS = ["#2b5c8f", "#d95f02"]


X, _ = make_moons(n_samples=500, noise=0.06, random_state=42)
model = CLARANS(n_clusters=2, num_local=5, random_state=42)
model.fit(X)

fig, ax = plt.subplots(figsize=(6.0, 4.5))
for k in range(2):
    mask = model.labels_ == k
    ax.scatter(X[mask, 0], X[mask, 1], c=COLORS[k], s=25, alpha=0.65, edgecolors="none")

centers = model.cluster_centers_
ax.scatter(
    centers[:, 0],
    centers[:, 1],
    marker="x",
    s=90,
    c="black",
    linewidths=1.8,
    label="Medoids",
)

ax.set_title("CLARANS on Interleaving Moons ($k=2$)", fontsize=11, pad=8)
ax.set_xticks([])
ax.set_yticks([])
for spine in ax.spines.values():
    spine.set_color("#cccccc")
    spine.set_linewidth(0.8)

ax.legend(loc="upper right", frameon=False, fontsize=9)
plt.tight_layout()
plt.show()