bsxfun slow?
in [Matlab]
|
Prev: Finding max of simple function
Next: Filter Design
From: Andreas Lobinger on 10 Jun 2010 02:33 Aloha, in the SW i'm working on, there are some operations which f.e. build a 'cross' difference, so from all elements from one vector to all elements in the second vector. d = repmat(a,1,size(b,2)) - repmat(b,size(a,1),1); As my a changes over the runtime and the b is constant, i store the possible repmat(b...) and build the repmat(a,...) in runtime. Recently i learned, that bsxfun(@minus, ...) would be an option not to do the repmat and have the difference calculated in place. However, the profiler tells me, that using bsxfun is slower than the initial solution (?). So if the initial solution needs something like 1.0 (time) for the difference (-) and around 0.6 for the repmat(b...) the bsxfun is 2.0, so slower then the sum of the operations. What i'm doing wrong? Wishing a happy day, LOBI
From: Walter Roberson on 10 Jun 2010 02:40 Andreas Lobinger wrote: > Recently i learned, that bsxfun(@minus, ...) would be an > option not to do the repmat and have the difference calculated > in place. > > However, the profiler tells me, that using bsxfun is slower > than the initial solution (?). Instead of @minus there, try 'minus' (the string 'minus') as bsxfun has built-in logic to handle that if I recall.
From: Oliver Woodford on 10 Jun 2010 04:59 Walter Roberson wrote: > Instead of @minus there, try 'minus' (the string 'minus') as bsxfun has > built-in logic to handle that if I recall. Apparently not: >> bsxfun('minus', [1; 2], [0 1]) ??? Error using ==> bsxfun First Argument must be a function handle. A pity. It's a good idea.
From: Peter Perkins on 10 Jun 2010 08:39 On 6/10/2010 2:33 AM, Andreas Lobinger wrote: > in the SW i'm working on, there are some operations which > f.e. build a 'cross' difference, so from all elements from > one vector to all elements in the second vector. > > d = repmat(a,1,size(b,2)) - repmat(b,size(a,1),1); > However, the profiler tells me, that using bsxfun is slower > than the initial solution (?). LOBI, how big are a and b?
From: Matt Fig on 10 Jun 2010 09:55
Walter Roberson <roberson(a)hushmail.com> wrote in message <6N%Pn.72624$HG1.14036(a)newsfe21.iad>... > Andreas Lobinger wrote: > > > Recently i learned, that bsxfun(@minus, ...) would be an > > option not to do the repmat and have the difference calculated > > in place. > > > > However, the profiler tells me, that using bsxfun is slower > > than the initial solution (?). > > Instead of @minus there, try 'minus' (the string 'minus') as bsxfun has > built-in logic to handle that if I recall. You might be thinking of the limited string inputs to CELLFUN, which are indeed much faster than using equivalent function handles. I wish TMW would increase the number of these for CELLFUN and add some for BSXFUN. |