Week 9 Flashcards
1
Q
Common metrics for dissimilarity for clustering
A
Euclidean distance
Squared euclidean distance
Manhattan distance
Cosine distance
2
Q
What is k-means clustering
A
Partitioning the input data into K>0 distinct clusters
3
Q
How does k-means work
A
Iterative algorithm, uses squared Euclidean distance and the centroid (average) as cluster representative.
Can only accurately model spherical clusters this way.
4
Q
is k-means good?
A
typically finds sub-optimal solutions as k-clustering problem is NP-hard
5
Q
What is hierarchical clustering
A
Generate a hierarchy of clusters with different resolutions
Computationally heavy
Agglomerative / Divisive
6
Q
How do we validate cluster quality?
A
Compactness (low variance) and separation (large spread)
