MATLAB Central Newsreader - vectorization suggestion

vectorization suggestion

Piero Lanucara — Sat, 16 Jul 2011 22:04:08 +0000

Hi to all,
There exist a simple and fast way to do this:

v_orig=[1 0 2 0 3 0 4 0 0 5 0 0 6] in
v_end=[1 2 2 3 3 4 4 5 5 5 6 6 6]

in the end the 0 values has to be replace with the first non zero element shifting towards the right direction....of course this vector is only an example because I use very huge vector like this!
thank's in advance
Piero

Re: vectorization suggestion

Roger Stafford — Sun, 17 Jul 2011 00:13:08 +0000

"Piero Lanucara" <lanucara@caspur.it> wrote in message <ivt1so$1uu$1@newscl01ah.mathworks.com>...
> Hi to all,
> There exist a simple and fast way to do this:
>
> v_orig=[1 0 2 0 3 0 4 0 0 5 0 0 6] in
> v_end=[1 2 2 3 3 4 4 5 5 5 6 6 6]
>
> in the end the 0 values has to be replace with the first non zero element shifting towards the right direction....of course this vector is only an example because I use very huge vector like this!
> thank's in advance
> Piero
- - - - - - - - - -
What's wrong with just using a simple for-loop? It may be about as efficient as any vectorization and is a lot easier to code.

v_end = v_orig;
t = 0;
for k = length(v_end):-1:1
if v_end(k) == 0, v_end(k) = t; else, t = v_end(k); end
end

Roger Stafford

Re: vectorization suggestion

Matt J — Sun, 17 Jul 2011 00:40:13 +0000

"Roger Stafford" wrote in message <ivt9ek$jq3$1@newscl01ah.mathworks.com>...
>
> What's wrong with just using a simple for-loop? It may be about as efficient as any vectorization and is a lot easier to code.
>
=========================

I agree that this might be a case of vectorization overkill. Nevertheless, the following might be a better alternative to a for-loop if the strings of zeros are all fairly short

v_end=v_orig;
idx=1;
while any(idx)

   s = circshift(v_end,[0,-1]);
   idx=s & ~v_end;
   v_end(idx)=s(idx);

end

Re: vectorization suggestion

Piero Lanucara — Sun, 17 Jul 2011 07:44:08 +0000

many thank's for the suggestions
I agree that probably the first implentation is faster with huge (the dimension is more or less the half number of non zeros of large symmetric sparse) matrix
probably there's some chance to do calculation in place so some benefit should occurr
thanks!

P.

"Matt J" wrote in message <ivtb1d$ng6$1@newscl01ah.mathworks.com>...
> "Roger Stafford" wrote in message <ivt9ek$jq3$1@newscl01ah.mathworks.com>...
> >
> > What's wrong with just using a simple for-loop? It may be about as efficient as any vectorization and is a lot easier to code.
> >
> =========================
>
> I agree that this might be a case of vectorization overkill. Nevertheless, the following might be a better alternative to a for-loop if the strings of zeros are all fairly short
>
> v_end=v_orig;
> idx=1;
> while any(idx)
>
> s = circshift(v_end,[0,-1]);
> idx=s & ~v_end;
> v_end(idx)=s(idx);
>
> end

Re: vectorization suggestion

Bruno Luong — Mon, 18 Jul 2011 06:23:08 +0000

v_orig=[0 1 0 2 0 3 0 4 0 0 5 0 0 6]

v = fliplr(v_orig);
idx = v~=0;
d = diff(v(idx));
idx(find(idx,1,'first')) = false;
v(idx) = d;
v = fliplr(cumsum(v))

% Bruno