Clever solution sought: Summing values from an array into a vector component given in a second array
1 view (last 30 days)
Show older comments
Matthew Masarik
on 22 Aug 2014
Edited: per isakson
on 22 Aug 2014
I have two arrays -- an array of indices and an array of values, both are K x M. The index array has values between 1:M. Now I want to obtain the vector whose jth component is sum( valArray( indArray == j ) ).
Can you think of way to do this that beats the simple for loop solution? I have a couple of bsxfun solutions to do the ( indArray == j ) in one shot, but those don't beat the for loop. I also have a way to do this using a sparse array, but that doesn't beat the for loop either.
Here's code to set up the problem and find the answer using a for loop:
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%NOTE: set M and K to be much larger for the real problem
M = 10;
K = 5;
indArray = randi(M,[K M]);
valArray = rand([K M]);
result = zeros(M,1);
for ii = 1 : M * K
temp = indArray(ii);
result(temp) = result(temp) + valArray(ii);
end
Thanks for taking a look.
0 Comments
Accepted Answer
More Answers (2)
Roger Stafford
on 22 Aug 2014
You might try using accumarray:
A = accumarray(indArray(:),valArray(:),[M,1]);
0 Comments
per isakson
on 22 Aug 2014
Edited: per isakson
on 22 Aug 2014
Yes, accumarray beats the loop for large arrays (R2013a)
M = 2e3;
K = 5e3;
ix = randi(M,[K,M]);
val = rand([K,M]);
tic
out = zeros(M,1);
for ii = 1 : M * K
out(ix(ii)) = out(ix(ii)) + val(ii);
end
toc
tic
acc = accumarray( ix(:), val(:), [], @sum );
toc
assert( all( abs(out-acc)<1e-6 ), 'A:B:C', 'Results differ' )
returns
Elapsed time is 0.330848 seconds.
Elapsed time is 0.179041 seconds.
 
However
M = 100;
K = 40;
returns
Elapsed time is 0.000125 seconds.
Elapsed time is 0.000258 seconds.
0 Comments
See Also
Categories
Find more on Logical in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!