Prev: Adapt function problem
Next: Help on implementing motor torque control reg with already available inefficient motor
From: Matt Fig on 2 Jul 2010 10:14
How about this one:

% Data
lamda = rand(100,3)-.5
[l,ind] = sort(abs(lamda),2);

% Engine
lamda3 = sign(lamda);
lamda3 = l .* lamda3(cumsum(ones(size(lamda3)))+(ind-1)*size(lamda3,1));
From: Bruno Luong on 2 Jul 2010 10:32

> % Engine
> lamda3 = sign(lamda);
> lamda3 = l .* lamda3(cumsum(ones(size(lamda3)))+(ind-1)*size(lamda3,1));

It seems the same work (on the sign array) can be done to extract the value directly, so I wonder what is the advantage of the above.

lamda4 = lamda(cumsum(ones(size(lamda)))+(ind-1)*size(lamda,1))

Bruno
From: Matt J on 2 Jul 2010 12:16
"Bruno Luong" <b.luong(a)fogale.findmycountry> wrote in message <i0kroc$cb2$1(a)fred.mathworks.com>...

> > If lamda is only nx3, I would actually expect a for-loop over columns, as opposed to rows, to be the most efficient approach, faster even than anything vectorized:
> >
> > for i=1:size(lamda,2)
> >
> > lamda2(:,i) = lamda(ind(:,i),i);
> >
> > end
>
> Nice try, but I believe that's won't work.

Hmmm. I guess not. Oh well, a double for-loop might still be pretty fast, as long as we're talking about 3 columns,

lamda2=lamda;
for i=1:3
for j=1:3

idx= (ind(:,i)==j);
lamda2(idx,i) = lamda(idx,j) ;

end
end
From: Matt Fig on 2 Jul 2010 13:07
"Bruno Luong" <b.luong(a)fogale.findmycountry> wrote in message <i0kt94$p8b$1(a)fred.mathworks.com>...
>
> > % Engine
> > lamda3 = sign(lamda);
> > lamda3 = l .* lamda3(cumsum(ones(size(lamda3)))+(ind-1)*size(lamda3,1));
>
> It seems the same work (on the sign array) can be done to extract the value directly, so I wonder what is the advantage of the above.
>
> lamda4 = lamda(cumsum(ones(size(lamda)))+(ind-1)*size(lamda,1))
>
> Bruno


Good point! I guess that is what 5 hours of sleep does to a person....
From: Thomas on 5 Jul 2010 09:23
> If you like treat this problem:
>
> lamda = rand(10,3)-0.5;
> [l,ind] = sort(abs(lamda),2);
> for i=1:size(lamda,1)
> lamda2(i,:) = lamda(i,ind(i,:));
> end
>
Here is another short vectorized solution

lamdaabs = abs(lamda);
lamdamax = max(lamdaabs ,[],2);
[m,loc] = ismember(lamdamax,lamdaabs);
lamda2 = lamda(loc);

Thomas

Thanks to all who engage this problem!
First  |  Prev  | 
Pages: 1 2 3
Prev: Adapt function problem
Next: Help on implementing motor torque control reg with already available inefficient motor