low speed of 'ismember' function

8 views (last 30 days)
Ehsan
Ehsan on 29 Aug 2014
Commented: Ehsan on 29 Aug 2014
hi. in my code there is a for loop like the following:
for i=1:n
N=find(~ismember(Vector1,Vector2))
do something...
end
This loop is repeated several thousand times. 'ismember' function is very slow. So that most of the runtime for 'ismember' function lost. Is there any alternative to this?

Answers (3)

Johan Löfberg
Johan Löfberg on 29 Aug 2014
If data is sorted, using ismembc could be a quick fix.
  2 Comments
Johan Löfberg
Johan Löfberg on 29 Aug 2014
Then I would scratch ismember and do sort+ismembc (ismember contains a lot of overhead. look into it and you will see that it basically does sort+ismembc once all data is checked etc)

Sign in to comment.


David Sanchez
David Sanchez on 29 Aug 2014
Depending on your data, you might try to code your own function avoiding some conditions and calls to sub-functions within the ismember function.
You can take a look at ismember code with:
type ismember
and try to make it easier.
  3 Comments
Ehsan
Ehsan on 29 Aug 2014
I just do not understand. What is your suggestion? Please change the code according to your suggestion.

Sign in to comment.


Titus Edelhofer
Titus Edelhofer on 29 Aug 2014
Hi,
does "Vector1" and "Vector2" changes in every iteration? Often you see something like
for i=1:n
v = ismember(x(i), y);
% do something
end
which can be changed to
vAll = ismember(x, y);
for i=1:n
% now do something with vAll(i)
end
Titus
  1 Comment
Ehsan
Ehsan on 29 Aug 2014
this is my code:
for k=1:nAnt
ant(k).Tour=randi([1,nCity],1,1);
end
for i=2:nCity
for k=1:nAnt
N=find(~ismember(1:nCity,ant(k).Tour));
% do somthing
end
end

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!