Collapsing preallocated memory of sparse matrix

Question

Hyeokjin Jho 2019-1-22

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/440854-collapsing-preallocated-memory-of-sparse-matrix

评论： Bjorn Gustavsson 2019-1-23

I used sparse matrix for some calculation, and I roughly know how many nnz will be stored. So I preallocated a slightly larger memory when I initialize that sparse matrix.

reserveRate = 1.2 % 120%
A = spalloc(n,m,estimated_nnz * reserveRate);

After finishing the calculation, since the memory has some unfilled margin, I would like to remove those space. Is there any way to do so?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Bjorn Gustavsson 2019-1-22

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/440854-collapsing-preallocated-memory-of-sparse-matrix#answer_357458

Is there a particular reason you dont collect the non-zero values and their matrix indices into 3 arrays and then after all those are completed first then generating the sparse array? That is the general advice to not over-allocate.

HTH

2 个评论
显示无隐藏无

Hyeokjin Jho 2019-1-23

It's a quite big data. The array takes memory over 50%, so duplicating it will lead to shortage of memory.

Bjorn Gustavsson 2019-1-23

But if your arrays with the non-zero values and the corresponding row and column indices take up over 50% of your memory wouldn't you rapidly run into problems as soon as you start doing operations on the sparse array?

Sad follow-up suggestion: read the indices first, then you candetermine the exact size needed so you can allocate the minimal A, and then do another read to fill it? If it is so that you just get the values and indices in some calculation maybe another way would be to store them in arrays, save to file(s) - then you might be able to wiggle your way around by determining the size of A and then populate it. I know these are "not pretty" suggestions...

请先登录，再进行评论。