Flat clustering (fcluster) in SciPy - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is flat clustering in the context of hierarchical clustering?

Flat clustering means cutting the hierarchical clustering tree at a certain level to form groups (clusters) without any nested structure.

beginner

What does the fcluster function in scipy.cluster.hierarchy do?

The fcluster function cuts a hierarchical clustering tree to assign cluster labels to data points based on a threshold or number of clusters.

intermediate

Which parameters are important when using fcluster?

Key parameters are:<br>• Z: linkage matrix from hierarchical clustering<br>• t: threshold to cut the tree<br>• criterion: method to decide clusters (e.g., 'distance', 'maxclust')

intermediate

How does the criterion='maxclust' option work in fcluster?

It forms a flat clustering with a maximum number of clusters specified by t. The tree is cut to get at most that many clusters.

beginner

Why is flat clustering useful after hierarchical clustering?

Flat clustering simplifies the hierarchical tree into clear groups, making it easier to analyze and use clusters for tasks like labeling or further analysis.

What does the fcluster function return?

A An array of cluster labels for each data point

B A linkage matrix

C A dendrogram plot

D The original data sorted

fcluster returns an array where each element is the cluster label assigned to the corresponding data point.

Which criterion in fcluster cuts the tree by distance threshold?

A 'maxclust'

B 'distance'

C 'inconsistent'

D 'monocrit'

The 'distance' criterion cuts the hierarchical tree at a specified distance threshold t.

If you want exactly 3 clusters from hierarchical clustering, which criterion should you use with fcluster?

A 'maxclust'

B 'distance'

C 'inconsistent'

D 'monocrit'

The 'maxclust' criterion cuts the tree to form a maximum number of clusters specified by t, so use t=3.

What input does fcluster require to assign clusters?

A Raw data points

B Cluster centroids

C Linkage matrix from hierarchical clustering

D Distance matrix only

fcluster needs the linkage matrix Z which encodes the hierarchical clustering structure.

Which of these is NOT a valid criterion for fcluster?

A 'inconsistent'

B 'maxclust'

C 'distance'

D 'kmeans'

'kmeans' is not a criterion for fcluster. It is a separate clustering method.

Explain how the fcluster function works and why it is used after hierarchical clustering.

Think about how to get flat groups from a tree structure.

Describe the difference between using criterion='distance' and criterion='maxclust' in fcluster.

One uses a distance limit, the other uses a cluster count.