kmeans clustering of matrices/arrays

I have a 1X52 cell. Each element of cell is a 33X35 matrix. Need to classify these matrices using kmeans. I tried kmeans(X,k). But it demands X to be a matrix and not a cell.

Is there any other way?

[Wayne King]

Are you trying to work with each matrix separately, or are you trying to cluster the large matrix -- (52*33,35)

assuming each row is an observation on 35 variables and the variables are the same for each matrix.

[pranesh]

yes i want to deal with each matrix separately. The command cell2mat works in the same way as you have mentioned.

Answer 1

[Image Analyst]

So make it a 3D array, which I would have done in the first place. Why mess with cells when you don't have to?

Anyway, since you have cells, just extract X from the cell and call kmeans

for c = 1 : length(myCellArray)
  X = myCellArray{c}; % Extract double array from cell.
  out = kmeans(X,k);
end

[pranesh]

thanks. I did that. But still unable to cluster the matrices. I am getting the following message.

??? Error using ==> kmeans>batchUpdate at 435 Empty cluster created at iteration 1.

Error in ==> kmeans at 336 converged = batchUpdate();

Error in ==> name at 51 out = kmeans(X,3);

[Walter Roberson]

That can happen by chance with random initialization. It can, though, also happen if you have a number of duplicate points.