sum from i to d over lambda i (W), W- average of cluster autocorrelation matrices, lambda i (W) - measure of the variance of the average cluster along one coordinate, tr(W) - measure for the complete variance of the average cluster
1 of 5
det(W)
product from i to d over lambda i (W) , measure for the volume of the average cluster, leads to custer with small volumes on average, not necessarily compactness
2 of 5
the third one
product from i to n over det(A i) ^|C i| , individual treatment of clusters avoids favoring an identical shape pf all clusters, problem: for d dimensions at least d+1 data per cluster are necessary (otherwise det = 0)
3 of 5
Maximize
tr(BW^-1) = tr((A-W)W^-1) = tr ( AW^-1 -1) , small tr(W) yields small clusters, large tr(B) yields big variance of cluster centers
4 of 5
whateffs
s
5 of 5
Other cards in this set
Card 2
Front
det(W)
Back
product from i to d over lambda i (W) , measure for the volume of the average cluster, leads to custer with small volumes on average, not necessarily compactness
Comments
No comments have yet been made