How to filter a matrix?

I'm clustering my data with the aim to produce a force directed graph. I've got the script running but the graphs are too connected and in need of filtering. As I produce the charts from the pdist function I want to filter out all but the top 10 values per row so that each node has no more than 10 edges in the resulting graph.

[M, Idx] = max(A,[],2)

gives me the max value and the location along the row but I'm unsure how to find the top 10 positions for each row or how to convert that back into an array of the same size as A

[Stephen23]

How would the output be "an array of the same size as A"? Surely by "filtering out" the lesser values you have made your matrix smaller. How can it be the same size?

[Lucas Black]

Hi Stephen,

Sorry if I used the wrong terms. By filter out I mean keep the top ten values but replace all the other valued with 0.

Answer 1

[Stephen23]

A = randi(9,6,20) % example matrix of integers
N = 10; % pick how many values to keep
[B,C] = sort(A,2);
B(:,1:end-N) = 0;
R = (1:size(B,1))'*ones(1,size(B,2));
Z = zeros(size(B));
Z(sub2ind(size(B),R,C)) = B % output

Where the example input and output matrix is:

A =
   5   6   7   4   6   5   5   9   3   2   5   7   5   2   1   8   4   9   5   4
   9   3   8   5   7   4   3   2   5   5   8   5   3   6   5   2   9   6   4   7
   5   9   5   9   3   8   3   5   2   3   8   4   2   9   1   5   4   6   5   9
   4   6   8   7   5   7   3   4   5   1   7   6   8   7   8   5   8   6   6   6
   3   2   7   5   6   1   8   4   8   6   2   2   5   8   5   9   9   1   5   2
   7   8   6   6   1   4   3   1   5   3   5   7   7   7   8   7   6   6   7   9
Z =
   0   6   7   0   6   0   0   9   0   0   5   7   5   0   0   8   0   9   5   0
   9   0   8   0   7   0   0   0   0   0   8   5   0   6   5   0   9   6   0   7
   0   9   0   9   0   8   0   5   0   0   8   0   0   9   0   5   0   6   5   9
   0   0   8   7   0   7   0   0   0   0   7   0   8   7   8   0   8   0   6   6
   0   0   7   0   6   0   8   0   8   6   0   0   0   8   5   9   9   0   5   0
   7   8   0   0   0   0   0   0   0   0   0   7   7   7   8   7   0   6   7   9

Note that picking the largest ten values may have unintended side effects: the code above picks the first of identical values, which means although one value might exist in multiple locations in one row, it is possible that only the first ones get returned (e.g 5 in the first row: there are actually six of them in the original matrix, but only three in the output). By "filtering" based on value rather than number of occurrences you could avoid this problem.

[Lucas Black]

Thanks Stephen, that worked a treat :D

Answer 2

[Azzi Abdelmalek]

A=randi(80,4,20)
n=size(A,1)
max1=zeros(n,10);
indices=zeros(n,10);
for k=1:size(A,1)
  [ii,jj]=sort(A(k,:),'descend');
  max1(k,:)=ii(1:10);
  indices(k,:)=jj(1:10);
end