MATLAB: Faster Method for removing duplicates

forfor loopMATLABoptimizationunique

I am trying to remove duplicate nodes from an auto mesh. I have attached a simple script. This works, but takes FOREVER if there are many (greater than 10,000) nodes. Does anyone have an alternate suggestion that would be faster?

node = 
[0.1234 5.6789;
 3.4567 9.8765;
 4.5678 8.7654;
 0.1234 5.6789]
% Remove duplicate nodes
node_array=cell(1,length(node));
for j = 1:length(node)
  node_array(j,1) = {node(j,:)};
end
node_array_cat = cat(1,node_array{:});
[node junk node_map] = unique(node_array_cat,'rows');
node = 
[0.1234 5.6789;
 3.4567 9.8765;
 4.5678 8.7654]

Best Answer

 >> node_u = unique(node, 'rows', 'stable')
 node_u =
    0.1234    5.6789
    3.4567    9.8765
    4.5678    8.7654

should be more efficient. What was the purpose of this conversion to cell array?

Related Solutions

MATLAB: How to filter column from a text file

fmt = '%-6d  %.3f  %.4f  %.4f  %.3f\n';
mat = dlmread('test_old.txt');
[fid,msg] = fopen('test_new.txt','wt');
assert(fid>=3,msg)
fprintf(fid,fmt,mat(:,[1,4:end]).')
fclose(fid);

Or simply use Notepad++: select those columns using shift+alt, then press delete.

MATLAB: Building a matrix starting from another matrix

At first I'd simplify the code. E.g. this is useless:

if (nodes(i,1) == nodes(j,1) <OR> nodes(i,1) ~= nodes(j,1))

Of course two number are either equal or not equal, as long as they are not NaNs.

function distances = net_dist(nodes,heads)
nnodes = size(nodes, 1);  % Better than "length(nodes(:,1))"
distances = zeros(nnodes);
for i = 1 : nnodes
    for j = 1 : nnodes
        match = ismember(nodes(j,1),heads(:,1));
        if nodes(i,1) == nodes(j,1) && match
            distances(i,j) = sqrt((nodes(i,2) - nodes(j,2))^2 + ...
                                  (nodes(i,3) - nodes(j,3))^2);
        elseif match
            distances(i,j) = sqrt((nodes(i,2) - nodes(j,2))^2 + ...
                                  (nodes(i,3) - nodes(j,3))^2); 
        else  % Now "elseif ~match" *must* be true!
            distances(i,j) = 1E9;
        end
    end
end

Now we see, that the first two IF-branches are performed whenever match is true. So a further simplification:

distances = repmat(1E9, nnodex, nnodes);
matchList = ismember(nodes(:,1), heads(:,1));
for i = 1 : nnodes
    for j = 1 : nnodes
        if matchList(j)
            distances(i,j) = sqrt((nodes(i,2) - nodes(j,2))^2 + ...
                                  (nodes(i,3) - nodes(j,3))^2);
        end
    end
end

Finally I do not understand also, why you get zeros in the output.

Best Answer

Related Solutions

MATLAB: How to filter column from a text file

MATLAB: Building a matrix starting from another matrix

Related Question