pointwise multiplication of sparse matrix with full array

2 次查看（过去 30 天）
Sandra Martinez
Sandra Martinez 2022-12-12
评论： Sandra Martinez 2022-12-13
Ran in:
How can I do this operation?
A=sparse([1 0;0 0]);
B=rand(2,2,2,2);
C=A.*B;
Error using .*
N-D sparse output arrays are not supported.
the only way is converting B into a sparse array with this function?
https://www.mathworks.com/matlabcentral/fileexchange/29832-n-dimensional-sparse-arrays?s_tid=srchtitle
remark: A is sparse but B is not.

请先登录，再回答此问题。

采纳的回答

Torsten
Torsten 2022-12-12
编辑：Torsten 2022-12-12
Ran in:
rand(2,2,2,2) generates a (2x2x2x2) array for B. You cannot multiply A with such a B elementwise because the array sizes don't match.
A=sparse([1 0;0 0]);
B=rand(2,2);
C=A.*B
C =
(1,1) 0.6805
Or what do you expect as result for B being (2x2x2x2) ?
  2 个评论
Torsten
Torsten 2022-12-12
@Sandra Martinez comment moved here:
Yes, I know. But if A is a full matrix this is possible ( in that case matlab makes the pointwisemultiplicación of B(: , : , i,j) with A, for each i,j) The problem here is with a sparse matix and pointwise multiplication

请先登录，再进行评论。

更多回答（1 个）

Sandra Martinez
Sandra Martinez 2022-12-13
Ran in:
Yes, this works now
M=rand(2,2,2,2);
N=[1 0;0 0];
A=ndSparse(M);%You need to download this function
Unrecognized function or variable 'ndSparse'.
B=sparse(N);
C=A.*B;
But, what do you mean by "artificial data structure"? In fact, I measure times and making pontiwise multiplication with full arrays takes less time (a factor 7).
n=50;
for i=1:20;
M=rand(n,n,n,n);
N=zeros(n,n);
N(1,1)=2;
tic
L=M.*N;%pointwise multiplication with full arrays
t1=toc;
tt(i)=t1;
A=ndSparse(M);%You need to download this function
B=sparse(N);
tic
C=A.*B;%pointwise multiplication using
sparse arrays
t2=toc;
tt2(i)=t2;
end
%%
% mean(tt)
% 0.0045
%mean(tt2)
% 0.0350
But if both are matices the time is the other way
for i=1:20;
M1=rand(n,n);
N1=zeros(n,n);
N(1,1)=2;
tic
L1=M1.*N1;
t3=toc;
tt3(i)=t3;
A1=ndSparse(M1);
B1=sparse(N1);
tic
C1=A1.*B1;
t4=toc;
tt4(i)=t4;
end
mean(tt4)/mean(tt3)
% 5.8859 doing pointwise multiplication using sprse matrices it is almost 6
% times faster.
Is there an explanation for that?
  2 个评论
Torsten
Torsten 2022-12-13
The author of ndSparse (Matt J) is active in this forum.
Maybe you should open a new question where ndSparse is part of the title and ask him directly.

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Performance and Memory 的更多信息

标签

产品


版本

R2021a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by