Select all (non-zero) columns of sparse matrix

Question

Quinten Rensen 2018-8-24

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/416106-select-all-non-zero-columns-of-sparse-matrix

评论： Quinten Rensen 2018-8-27

Hi all,

Since I'm working with very large and sparse matrices I use sparse() to define these matrices. I do this by defining the index of the rows and columns and their corresponding value in separate vectors. For example, for a row i and columns j,j-1,j+1, the values in the matrix are defined according:

s = 1;
row(s) = i; column(s)=j; K(s) = 1; s=s+1;
row(s) = i; column(s)=j-1; K(s) = -1; s=s+1;
row(s) = i; column(s)=j+1; K(s) = -1; s=s+1;
%This is done for all non-zero points
K_matrix = sparse(row,column,K);

This works great when specifying distinct values. However, how can I use this sparse indexing when the value depends on values in other rows of the matrix. For example:

K(i,:) = K(i-1,:) + K(i+1,:);

i.e. I need all the values of another row in order to construct the current row, since I can't do something like this:

row(s) = i; column(s)=:; K(s) = K(i-1,:)+K(i+1,:); s=s+1;

I use this way of defining the sparse matrix since it is recommended for speed. Does anyone know how I can circumvent this problem without Matlab becoming slow when building the matrix?

Kind regards

3 个评论
显示 1更早的评论隐藏 1更早的评论

Quinten Rensen 2018-8-26

I'm sorry, you can do that, but you will get the warning 'this sparse indexing expression is likely to be slow'. I introduced the row(s), column(s) indexing for speed (which I need, since the matrices become quite large).

Jan 2018-8-26

@Quinten Rensen: Posting an answer requires to understand, what you want to achieve, so please explain this. The purpose of "column(s)=:" is still not clear and it is no valid Matlab syntax.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Matt J 2018-8-26

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/416106-select-all-non-zero-columns-of-sparse-matrix#answer_334151

编辑：Matt J 2018-8-26

在 MATLAB Online 中打开

The best way will depend on the specifics of which rows are independent and how the remaining rows are derived from them. However, often the way would be to fill all of the independent rows first and then use vectorized operations to fill the remaining ones simultaneously. For example, starting with the sparse matrix,

>> K=spdiags([1 0 1 0 1].',0,5,5); full(K)
ans =
       1     0     0     0     0
       0     0     0     0     0
       0     0     1     0     0
       0     0     0     0     0
       0     0     0     0     1

I can fill in the even-numbered rows with the sum of their neighboring rows in a single statement,

>> K(2:2:end,:) = K(1:2:end-1,:)+K(3:2:end,:); full(K)
ans =
       1     0     0     0     0
       1     0     1     0     0
       0     0     1     0     0
       0     0     1     0     1
       0     0     0     0     1