Hierarchical Flashcards
1
Q
Hierarchical
A
No need to assume any particular number of clusters
Agglomerative Hierarchical
Divisive
2
Q
Algorithm
A
Compute proximity matrix Let each data point be a cluser Repeat Merge two closest Update distance matrix Until only a single cluster remains
3
Q
Complexity
A
O(N*N) proximity matrix
O(NNN) time in many cases
4
Q
Distance
A
Single Link or MIN
Complete Link or MAX
Mean Distance
Group Average
5
Q
Number of Clusters
A
Internal Criteria derived from data itself BSS-WSS Cohesion evaluates how similar are the points in same cluster Separation how far apart are points in different clusters External Prior or expert knowledge Relative Compare different solutions
6
Q
Cohesion
A
WSS = SUM(SUM(d(xj,ui)^2))
7
Q
Separation
A
BSS = SUM(|Ci|*d(u,ui)^2)
