It is needed to make FCM starts with an initial cluster value. Normally, FCM starts with input data to cluster and the number of clusters.
The command for fuzzy c mean clustering is FCM. genfis3 uses FCM to build a fuzzy inference system.
@ Mr. Awad
I do not think that one can define the initial guesses for center of clusters in matlab built-in function FCM. You can run the FCM a number of times and save the best result.
In Matlab (R2012b) centers are determined on the basis of membership functions:
stepfcn.m:
mf = U.^expo; % MF matrix after exponential modification
center = mf*data./((ones(size(data, 2), 1)*sum(mf'))'); % new center
Initial values of U matrix are random:
fcn.m:
U = initfcm(cluster_n, data_n); % Initial fuzzy partition
initfcm.m:
U = rand(cluster_n, data_n);
col_sum = sum(U);
U = U./col_sum(ones(cluster_n, 1), :);
So you can modify stepfcn.m:
%adding last parameter iter
function [U_new, center, obj_fcn] = stepfcm(data, U, cluster_n, expo, iter)
...
mf = U.^expo; % MF matrix after exponential modification
if (iter ==1)
center = [ your data];
else
center = mf*data./((ones(size(data, 2), 1)*sum(mf'))'); % new center
end;
...
And modify fcm.m:
...
[U, center, obj_fcn(i)] = stepfcm(data, U, cluster_n, expo, i);
...
Good luck!
You could the code attached. A simple code of FCM for Matlab/Octave as a function. You could change the initial coordinates of cluster centers as you see fit.
good luck.
- Badges
- Science topic