telisman
Computer
- Feb 25, 2007
- 1
When MATLAB kmeans implementation calculates values of matrix 'Del'
that will be used as "reassignment criterion", in the case of squared
euclidian distance it kind of "weights" every element dij (i = 1, 2,
... n , j = 1, 2, ... k) of the 'Del' matrix with tha ratio of
(initial members of cluster j / members of cluster j if data point j
is moved from -or to- cluster j).
I am referring to the m(i)./(m(i) + sgn) factor! What is the purpose
of these 'weights'? Why doesn it use only the plain squared
distances?
Any help is greatly appreciated.
Thanx in advance.
PS: I am trying to reverse engineer the kmeans matlab code because I need to run kmeans with my own custom distance function. If anyone has any solution on this issue too, pleas help.
Thanx again
that will be used as "reassignment criterion", in the case of squared
euclidian distance it kind of "weights" every element dij (i = 1, 2,
... n , j = 1, 2, ... k) of the 'Del' matrix with tha ratio of
(initial members of cluster j / members of cluster j if data point j
is moved from -or to- cluster j).
I am referring to the m(i)./(m(i) + sgn) factor! What is the purpose
of these 'weights'? Why doesn it use only the plain squared
distances?
Any help is greatly appreciated.
Thanx in advance.
PS: I am trying to reverse engineer the kmeans matlab code because I need to run kmeans with my own custom distance function. If anyone has any solution on this issue too, pleas help.
Thanx again
