Thread Subject:

Replacing for loops

I have been attempting to write a programme to run through all possible combinations of elements from two lists and add the results of a calculation on each pair together. the following is my current code, which works:

function ans = cor(train1, train2, tau)
ans = 0;
length1 = numel(train1);
length2 = numel(train2);
for n=1:length1
    for m= 1:length2
        ans = ans + exp(-abs(train1(n) - train2(m))/tau);
    end
end
end

tau is a scalar, while train1 and train2 are vectors of variable length. Can anyone give me a tip as to how to replace the for loops with something that will work faster?

I'm quite new to MATLAB, so I apologise if it's something trivially easy that I've missed.

"Charles Dillon" wrote in message <it7vvt$es8$1@newscl01ah.mathworks.com>...
> I have been attempting to write a programme to run through all possible combinations of elements from two lists and add the results of a calculation on each pair together. the following is my current code, which works:
>
> function ans = cor(train1, train2, tau)
> ans = 0;
> length1 = numel(train1);
> length2 = numel(train2);
> for n=1:length1
> for m= 1:length2
> ans = ans + exp(-abs(train1(n) - train2(m))/tau);
> end
> end
> end
>
> tau is a scalar, while train1 and train2 are vectors of variable length. Can anyone give me a tip as to how to replace the for loops with something that will work faster?
>
> I'm quite new to MATLAB, so I apologise if it's something trivially easy that I've missed.


I can't get rid of both loops, but can offer a suggestion to get rid of one of the loops:

%%%%%%%%%%%%%%%%
function ans = cor(train1, train2, tau)
ans = 0;
 for n=1:numel(train1) % length1 assignment unnecessary
   ans = ans + sum(exp(-abs(train1(n) - train2)/tau));
 end % for loop

end % function loop

%%%%%%%%%%%%%%

ans + exp(-abs(train1(n) - train2)/tau); % this will give you a [1 x length(train2)] matrix, so I've added the "sum" command to add them all up and then add that value to the "ans" value.

Help any?

A bit off topic but you should avoid using the variable ans. It is already
used by MATLAB for the results to calculations if they aren't assigned to a
variable. For instance:

2+4;

will yield ans=6



"Charles Dillon" <charlesd@live.ie> wrote in message
news:it7vvt$es8$1@newscl01ah.mathworks.com...
> I have been attempting to write a programme to run through all possible
> combinations of elements from two lists and add the results of a
> calculation on each pair together. the following is my current code, which
> works:
>
> function ans = cor(train1, train2, tau)
> ans = 0;
> length1 = numel(train1);
> length2 = numel(train2);
> for n=1:length1
> for m= 1:length2
> ans = ans + exp(-abs(train1(n) - train2(m))/tau);
> end
> end
> end
>
> tau is a scalar, while train1 and train2 are vectors of variable length.
> Can anyone give me a tip as to how to replace the for loops with something
> that will work faster?
>
> I'm quite new to MATLAB, so I apologise if it's something trivially easy
> that I've missed.

"Charles Dillon" wrote in message <it7vvt$es8$1@newscl01ah.mathworks.com>...
> ans = 0;
> length1 = numel(train1);
> length2 = numel(train2);
> for n=1:length1
> for m= 1:length2
> ans = ans + exp(-abs(train1(n) - train2(m))/tau);
> end
> end
> end
- - - - - - - - -
  You can use the 'bsxfun' for this. I will assume that train1 and train2 are both column vectors or both row vectors.

 a = sum(sum(exp(-abs(bsxfun(@minus,train1/tau,train2.'/tau))))):

  As Shawn says, don't use 'ans' for the name of a variable. It is a reserved matlab name.

Roger Stafford

"Corey Scheip" wrote in message <it815a$ij2$1@newscl01ah.mathworks.com>...
> "Charles Dillon" wrote in message <it7vvt$es8$1@newscl01ah.mathworks.com>...
> > I have been attempting to write a programme to run through all possible combinations of elements from two lists and add the results of a calculation on each pair together. the following is my current code, which works:
> >
> > function ans = cor(train1, train2, tau)
> > ans = 0;
> > length1 = numel(train1);
> > length2 = numel(train2);
> > for n=1:length1
> > for m= 1:length2
> > ans = ans + exp(-abs(train1(n) - train2(m))/tau);
> > end
> > end
> > end
> >
> > tau is a scalar, while train1 and train2 are vectors of variable length. Can anyone give me a tip as to how to replace the for loops with something that will work faster?
> >
> > I'm quite new to MATLAB, so I apologise if it's something trivially easy that I've missed.
>
>
> I can't get rid of both loops, but can offer a suggestion to get rid of one of the loops:
>
> %%%%%%%%%%%%%%%%
> function ans = cor(train1, train2, tau)
> ans = 0;
> for n=1:numel(train1) % length1 assignment unnecessary
> ans = ans + sum(exp(-abs(train1(n) - train2)/tau));
> end % for loop
>
> end % function loop
>
> %%%%%%%%%%%%%%
>
> ans + exp(-abs(train1(n) - train2)/tau); % this will give you a [1 x length(train2)] matrix, so I've added the "sum" command to add them all up and then add that value to the "ans" value.
>
> Help any?
Thanks for the help! I ended up going with Roger's suggestion, but thank you for the contribution nonetheless. My reason for the length1 assignment was that I thought that calling the numel function on every iteration of the loop would take up time. A lecturer told me that it was good practice, though he was referring to C++ at the time. Thanks again!

"Shawn Bonneau" <sbonneau@mathworks.com> wrote in message <it820l$knj$1@newscl01ah.mathworks.com>...
> A bit off topic but you should avoid using the variable ans. It is already
> used by MATLAB for the results to calculations if they aren't assigned to a
> variable. For instance:
>
> 2+4;
>
> will yield ans=6
>
>
>
> "Charles Dillon" <charlesd@live.ie> wrote in message
> news:it7vvt$es8$1@newscl01ah.mathworks.com...
> > I have been attempting to write a programme to run through all possible
> > combinations of elements from two lists and add the results of a
> > calculation on each pair together. the following is my current code, which
> > works:
> >
> > function ans = cor(train1, train2, tau)
> > ans = 0;
> > length1 = numel(train1);
> > length2 = numel(train2);
> > for n=1:length1
> > for m= 1:length2
> > ans = ans + exp(-abs(train1(n) - train2(m))/tau);
> > end
> > end
> > end
> >
> > tau is a scalar, while train1 and train2 are vectors of variable length.
> > Can anyone give me a tip as to how to replace the for loops with something
> > that will work faster?
> >
> > I'm quite new to MATLAB, so I apologise if it's something trivially easy
> > that I've missed.
Thanks for the advice. I hadn't even thought about the fact that I was overloading it. I'll bear that in mind.

"Roger Stafford" wrote in message <it85a0$1mp$1@newscl01ah.mathworks.com>...
> "Charles Dillon" wrote in message <it7vvt$es8$1@newscl01ah.mathworks.com>...
> > ans = 0;
> > length1 = numel(train1);
> > length2 = numel(train2);
> > for n=1:length1
> > for m= 1:length2
> > ans = ans + exp(-abs(train1(n) - train2(m))/tau);
> > end
> > end
> > end
> - - - - - - - - -
> You can use the 'bsxfun' for this. I will assume that train1 and train2 are both column vectors or both row vectors.
>
> a = sum(sum(exp(-abs(bsxfun(@minus,train1/tau,train2.'/tau))))):
>
> As Shawn says, don't use 'ans' for the name of a variable. It is a reserved matlab name.
>
> Roger Stafford
Thanks very much! That works perfectly! Much appreciated. I'll remember not to use ans in future.