Sparse building seems slow

Nick Henscheid

2011 6 17

0 个回答

9 次查看（30 天）

0 个投票

Hi, I'm working with some large scale LP matrices, exactly two nonzero entries per row and 500k+ rows. I'm using sparse to generate the matrices but it seems slow, about 20min for ~500k rows, 1hr+ for ~1M rows. This accounts for about 80% of my overall run time, once it hits the LP solver things go quickly. Any ideas on how to speed things up? I'm hoping to increase my number of data points which would require a larger matrix (~5M rows). Code below:

%row indices%
i=[1:n_active,1:n_active];
%column indices%
j=[row,col+n];
%constraint ones%
s=ones(2*n_active,1);
%filling sparse matrix% 
A=sparse(i,j,s,n_active,n+m,2*n_active);

I'm running on a modestly powered laptop (i5 dual core, 4g Ram). If I can provide any more info let me know, or feel free to tell me to just be patient =) Thanks! Nick

2 个评论
显示无隐藏无

Nick Henscheid 2011-6-28

I figured I'd give this a bump - any thoughts?

James 2011-11-15

I am working on the finite element method and I have the same problem. sparse is very slow when the matrix is in the order of 1M*1M. For FEM, this kind of matrices is very normal. It seems that sparse is the best code that Matlab can give. Perheps c or fortran is better?

请先登录，再进行评论。

请先登录，再回答此问题。

Follow Question