Why advanced clustering finds complex structures in ML Python - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why does DBSCAN find clusters of arbitrary shape?

DBSCAN is an advanced clustering algorithm. Why can it find clusters with complex shapes?

A Because it groups points based on density, not distance to a center

B Because it uses the mean of points to define cluster centers

C Because it requires the number of clusters as input

D Because it only works on spherical clusters

💡 Hint

Think about how DBSCAN decides which points belong together.

DBSCAN groups points that are close together in dense regions, allowing it to find clusters of any shape, unlike methods that rely on distance to a center.

❓ Predict Output

intermediate

Output of spectral clustering on a two-moon dataset

What is the shape of the clusters found by spectral clustering on a two-moon shaped dataset?

ML Python

from sklearn.datasets import make_moons
from sklearn.cluster import SpectralClustering
import numpy as np

X, _ = make_moons(n_samples=200, noise=0.05)
model = SpectralClustering(n_clusters=2, affinity='nearest_neighbors', assign_labels='kmeans')
labels = model.fit_predict(X)
unique_labels = np.unique(labels)
len(unique_labels)

A 0

B 2

C 3

D 1

💡 Hint

How many clusters does the model try to find?

Spectral clustering correctly finds 2 clusters matching the two moons shape.

❓ Model Choice

advanced

Best clustering model for non-convex clusters

You have data with clusters shaped like spirals. Which clustering model is best to find these complex shapes?

A Gaussian Mixture Model

B K-Means clustering

C DBSCAN

D Hierarchical clustering with single linkage

💡 Hint

Think about which method can find clusters based on density rather than shape.

DBSCAN can find clusters of arbitrary shape by grouping dense regions, making it suitable for spiral clusters.

❓ Hyperparameter

advanced

Effect of epsilon in DBSCAN clustering

In DBSCAN, what happens if you set the epsilon (eps) parameter too high?

A More clusters are found with smaller sizes

B Clusters become more separated and distinct

C The algorithm runs faster but finds no clusters

D Clusters merge and fewer clusters are found

💡 Hint

Think about how increasing the neighborhood radius affects grouping.

Setting eps too high causes points to be considered neighbors over large distances, merging clusters and reducing their number.

❓ Metrics

expert

Choosing the best metric to evaluate complex clustering

You applied an advanced clustering algorithm that finds complex-shaped clusters. Which metric best evaluates the quality of these clusters without knowing true labels?

A Silhouette score

B Accuracy

C Mean squared error

D Cross-entropy loss

💡 Hint

Consider metrics that measure cluster cohesion and separation without labels.

Silhouette score measures how well clusters are separated and how tight they are, useful when true labels are unknown.