Sparse Matrix Addition

Question

I am using a sparse matrix format implemented in scipy as csr_matrix. I have a mat variable which is in csr_matrix format and all its elements are non-negative. However, when I use mat + mat operation, the non-zero element number decreases which is quite strange to me. What want is a element-wise addition but why the non-element number will decreases as each of the element is non-negative.

Best Regards

Your question lacks detail, so in my answer I had to make assumptions. I recommend editing your question to add code demonstrating your problem. That will likely get you better answers. — David Alber

David Alber David Alber · Accepted Answer · 2012-07-06T06:01:21

The nnz member of csr_matrix in SciPy counts explicit zeros, so depending on how you create your matrix, this may explain what you are observing. You can see this behavior by explicitly setting zeros in a matrix.

>>> from scipy.sparse import csr_matrix
>>> A = csr_matrix((5, 5))
>>> A.nnz
0
>>> A[0, 0] = 0
>>> A.nnz
1
>>> A[1,1] = 0
>>> A.nnz
2

Now when you do an operation that creates a new matrix (such as matrix addition), the explicit zeros are not retained.

>>> B = A + A
>>> B.nnz
0

Sparse Matrix Addition

2 Answers