Splitting matrix and finding mean after each split

This is a follow up to a previous question that I posted.

My actual data is 12x416 matrix. Splitting this data across the median is not a better choice for me. I have to split it across the mean, even if the split is un-even.

So for each row, I find the first mean and then split it in two vectors. Then I find the mean for each vector and again split it. This continues until I have (say) 7 means for each row.

I wrote the following code for 1 row.

% v is a 12x416 matrix

vect1 = v(1, :);
m(1) = mean(vect1)
vect11 = vect1(vect1>m)
vect12 = vect1(vect1<m);
m(2) = mean(vect11);
m(3) = mean(vect12);
vect111 = vect11(vect11>m(2));
vect112 = vect11(vect11<m(2));
vect121 = vect12(vect12>m(3));
vect122 = vect12(vect12<m(3));
m(4) = mean(vect111);
m(5) = mean(vect112);
m(6) = mean(vect121);
m(7) = mean(vect122);

Is there an efficient way to write this all in a loop?

Thanks.

[Walter Roberson]

Note that you will need to write to a cell array rather than to a numeric matrix, because of the possibility of uneven splits.

[Walter Roberson]

What do you want the output data structure to look like? Do you need all of the intermediate splits, or just the final?

Is it certain than the array is sorted?

[M]

At the end, I need only those 7 means.

Initially the matrix is not sorted. But I can sort it if needed.

Answer 1

[Image Analyst]

I'm not sure what you're after but splitting data up into smaller and finer chunks is what quadtree decomposition does. If you have the Image Processing Toolbox, you might take a look at qtdecomp() and see if that does something like what you're trying to do.

[M]

What I am looking for is something very similar to k-means clustering. There is a kmeans() in Matlab but it is not exactly what I am looking for. The algorithm is slightly different.