Thread Subject: Vectorizing Double For Loops

So I am having trouble vectorizing double for loops so they are more efficient...

For example...

for yy=1:n
    for uu=1:n
        X(yy,uu)=Y(yy,uu)*W(yy,1)*Constant;
    end
end

the problem is multiplying by matrix dimensions that are not the same. thanks for your help

Your double For loop can be written as:

X = bsxfun(@times,Y,W(:,1))*Constant

and for large arrays will certainly be faster than a double For loop.

"omegayen " <omegayen@ameritech.net> wrote in message <h4ttp3$acv$1@fred.mathworks.com>...
> So I am having trouble vectorizing double for loops so they are more efficient...
>
> For example...
>
> for yy=1:n
> for uu=1:n
> X(yy,uu)=Y(yy,uu)*W(yy,1)*Constant;
> end
> end
>

X = Y .* repmat(W(:),1,size(Y,2)) * Constant

% OR

X = Constant * bsxfun(@times,Y,W(:))

% OR

X = Constant * (diag(W(:)) * Y)

% OR

X = diag(Constant * W(:) )* Y

% Bruno

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <h4u1m5$dm2$1@fred.mathworks.com>...
> "omegayen " <omegayen@ameritech.net> wrote in message <h4ttp3$acv$1@fred.mathworks.com>...
> > So I am having trouble vectorizing double for loops so they are more efficient...
> >
> > For example...
> >
> > for yy=1:n
> > for uu=1:n
> > X(yy,uu)=Y(yy,uu)*W(yy,1)*Constant;
> > end
> > end
> >
>
> X = Y .* repmat(W(:),1,size(Y,2)) * Constant
>
> % OR
>
> X = Constant * bsxfun(@times,Y,W(:))
>
> % OR
>
> X = Constant * (diag(W(:)) * Y)
>
> % OR
>
> X = diag(Constant * W(:) )* Y
>
> % Bruno

thanks Bruno that worked great..... perhaps you can help me with another, I can't figure out how to do it for some reason...

I want to compute the distance between points..... using 2 column vectors and ending with a square distance matrix

essentially

x= (column vector length n)
y= (column vector length n)

for i=1:n
    for j=1:n
        distance = sqrt ((x(i) - x(j)).^2 + (y(i) - y(j)).^2)
    end
end

the above implementation works, but I am trying to speed it up somehow, any thoughts would be welcomed

"omegayen " <omegayen@ameritech.net> wrote in message <h520qu$eo4$1@fred.mathworks.com>...

>
> essentially
>
> x= (column vector length n)
> y= (column vector length n)
>
> for i=1:n
> for j=1:n
> distance = sqrt ((x(i) - x(j)).^2 + (y(i) - y(j)).^2)
> end
> end
>
> the above implementation works, but I am trying to speed it up somehow, any thoughts would be welcomed

This should do:

dx = bsxfun(@minus,x(:),x(:).');
dy = bsxfun(@minus,y(:),y(:).');
distance = sqrt(dx.^2+dy.^2)

% Bruno

"Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <h522l2$745$1@fred.mathworks.com>...
> "omegayen " <omegayen@ameritech.net> wrote in message <h520qu$eo4$1@fred.mathworks.com>...
>
> >
> > essentially
> >
> > x= (column vector length n)
> > y= (column vector length n)
> >
> > for i=1:n
> > for j=1:n
> > distance = sqrt ((x(i) - x(j)).^2 + (y(i) - y(j)).^2)
> > end
> > end
> >
> > the above implementation works, but I am trying to speed it up somehow, any thoughts would be welcomed
>
> This should do:
>
> dx = bsxfun(@minus,x(:),x(:).');
> dy = bsxfun(@minus,y(:),y(:).');
> distance = sqrt(dx.^2+dy.^2)
>
> % Bruno

I forgot to specify that I can't use bsxfun with class sym (I am trying to use class sym in the implementation) so the method works great but bsxfun doesnt work with class sym

This is faster than the bsxfun soln anyway for n = 300.

distance = zeros(n,n); % Pre-allocate, Pre-allocate, Pre-allocate!
for ii=1:n
    for jj=ii:n % No need to start at 1 and duplicate calculations.
        distance(ii,jj) = sqrt((x(ii) - x(jj)).^2 + (y(ii) - y(jj)).^2);
    end
end
distance = distance + distance.'; % Do you really need the duplicate data (this step)?

"Matt Fig" <spamanon@yahoo.com> wrote in message <h524db$pp3$1@fred.mathworks.com>...
> This is faster than the bsxfun soln anyway for n = 300.
>
> distance = zeros(n,n); % Pre-allocate, Pre-allocate, Pre-allocate!
> for ii=1:n
> for jj=ii:n % No need to start at 1 and duplicate calculations.
> distance(ii,jj) = sqrt((x(ii) - x(jj)).^2 + (y(ii) - y(jj)).^2);
> end
> end
> distance = distance + distance.'; % Do you really need the duplicate data (this step)?

That should speed it up Matt, thanks....

does anyone know any implementation of Bruno's suggestion for class sym.....

basically can you multiply a column vector of length n by another column vector of length n to create a square matrix n x n using class sym?

"omegayen " <omegayen@ameritech.net> wrote in message <h535bi$2hl$1@fred.mathworks.com>...
> "Matt Fig" <spamanon@yahoo.com> wrote in message <h524db$pp3$1@fred.mathworks.com>...
> > This is faster than the bsxfun soln anyway for n = 300.
> >
> > distance = zeros(n,n); % Pre-allocate, Pre-allocate, Pre-allocate!
> > for ii=1:n
> > for jj=ii:n % No need to start at 1 and duplicate calculations.
> > distance(ii,jj) = sqrt((x(ii) - x(jj)).^2 + (y(ii) - y(jj)).^2);
> > end
> > end
> > distance = distance + distance.'; % Do you really need the duplicate data (this step)?
>
> That should speed it up Matt, thanks....
>
> does anyone know any implementation of Bruno's suggestion for class sym.....
>
> basically can you multiply a column vector of length n by another column vector of length n to create a square matrix n x n using class sym?

still wondering if anyone knows how to implement this without double for loops and for class sym, thanks