about scalar and Sparse Matrix multiplication
in [Matlab]
|
Prev: finding the four corners pixel location values of a small reectangular image in a large image
Next: Regarding State Flow....
From: Shiyuan Gu on 19 Jan 2010 00:40 I met a strange problem about the scalar and Sparse Matrix Multiplication. The following expression: v=0.25*M*w is equivalent to v=M*w; v=0.25*v; where M is a sparse matrix However, the latter is much faster way than the former way (about ten time). Why? Thanks
From: James Tursa on 19 Jan 2010 01:45 "Shiyuan Gu" <gshy2014(a)gmail.com> wrote in message <hj3gjm$9p5$1(a)fred.mathworks.com>... > I met a strange problem about the scalar and Sparse Matrix Multiplication. > The following expression: > v=0.25*M*w > is equivalent to > v=M*w; > v=0.25*v; > where M is a sparse matrix > However, > the latter is much faster way than the former way (about ten time). Why? > Thanks Depends on what w is, what the sizes of the variables are, and what MATLAB version you are running. If you do the M*w first then the result could be a lot smaller than M, so the subsequent multiply by 0.25 is not as much as doing the 0.25 multiply first. Also, different versions of MATLAB handle the scalar * sparse multiply differently, so you can get significantly different timing answers on different machines with different versions of MATLAB. James Tursa
From: Bruno Luong on 19 Jan 2010 02:35
"Shiyuan Gu" <gshy2014(a)gmail.com> wrote in message <hj3gjm$9p5$1(a)fred.mathworks.com>... > I met a strange problem about the scalar and Sparse Matrix Multiplication. > The following expression: > v=0.25*M*w > is equivalent to > v=M*w; > v=0.25*v; > where M is a sparse matrix > However, > the latter is much faster way than the former way (about ten time). Why? > Thanks The first version does this v = (0.25*M)*w The reason why it's slower is now obvious: in one case the scalar is multiplied with a matrix; in other case the scalar is multiplied with a vector. Bruno |