Question about vectorizing a search function

I have a Nx2 matrix, and I want to check for the existence of elements in the second column in my first column.

What I have set up is a binary tree, with a Nx2 matrix of point indices indicating which point is connected to what. (i.e.[1 2] means point 1 is connected to point 2, etc). For my terminal end points, they never exist as beginning points, only end points, so with the way I have it set up, they will never be found in the first column.

What I want to use is the find function, which is fast. I have something working with the ismember function, but that can get to be very slow with larger trees. Some sample code I have is:

   terminalIndices = ismember(connMx(:,2), connMx(:,1))

where connMx is my Nx2 matrix of point connection indices.

Other than that, what I would do (and saves some time with overlarge trees, but is too verbose) is looping from 1 to nPoints with the code:

   counter = 1;
   for i=1:nPoints
      a = find(ccoFaceMx(:,1) == i)
      if a ~= []
         termIdxArray(counter) = a
         counter = counter+1;
      end
   end

Again, for smaller trees this is fine, but I can at times have trees larger than 1000 elements, and this function begins to slow down considerably then.

Are there any alternatives to this? Thanks for your advice!

Answer 1

[Andrei Bobrov]

[a,b] = ismember(ccoFaceMx(:,1),1:nPoints);
[a1,c,c] = unique(b(a));
d = (1:numel(a))';
out = [num2cell(a1), accumarray(c,d(a),[],@(x){x})];

[Brian]

Hello,

This worked pretty well, but it seemed to do the exact opposite of what I wanted: it gave me entries that existed in the first column, when I wanted those that did not.

Still, though, thanks, this seems to be an improvement over what I had before.