CODEWITHSUNDEEP
pythonscipysparse-matrix
Vineeth Bhaskara
Vineeth Bhaskara

Reputation: 205

Scipy CSR Matrix Element-wise Addition

Unlike numpy arrays/matrices, CSR matrix seems not to allow automatic broadcasting. There are methods in the CSR implementation for element-wise multiplication, but not addition. How to add to a CSR Sparse matrix by a scalar efficiently?

Upvotes: 6

Views: 7222

Answers (1)

gboffi
gboffi

Reputation: 25023

Here we want to add a scalar to the non-zero entries and leave alone the matrix sparseness, i.e. do not touch the zero entries.

From the fine Scipy docs (** emphasis ** is mine):

Attributes

nnz                   Get the count of explicitly-stored values (nonzeros)  
has_sorted_indices    Determine whether the matrix has sorted indices  
dtype (dtype)         Data type of the matrix  
shape (2-tuple)       Shape of the matrix  
ndim  (int)           Number of dimensions (this is always 2)  
**data                CSR format data array of the matrix** 
indices               CSR format index array of the matrix  
indptr                CSR format index pointer array of the matrix

So I tried (the first part is "stolen" from the referenced documentation)

In [18]: from scipy import *

In [19]: from scipy.sparse import *

In [20]: row = array([0,0,1,2,2,2])
    ...: col = array([0,2,2,0,1,2])
    ...: data =array([1,2,3,4,5,6])
    ...: a = csr_matrix( (data,(row,col)), shape=(3,3))
    ...: 

In [21]: a.todense()
Out[21]: 
matrix([[1, 0, 2],
        [0, 0, 3],
        [4, 5, 6]], dtype=int64)

In [22]: a.data += 10

In [23]: a.todense()
Out[23]: 
matrix([[11,  0, 12],
        [ 0,  0, 13],
        [14, 15, 16]], dtype=int64)

In [24]:

It works. Should you save the original matrix you can use the constructor using a modified data array.

Disclaimer

This answer addresses this interpretation of the question

I have a sparse matrix, I want to add a scalar to the non zero entries, preserving the sparseness both of the matrix and of its programmatic representation.

My reasoning for choosing this interpretation is that adding a scalar to all entries turns the sparse matrix in a VERY dense matrix...

If this is the correct interpretation, I don't know: on one hand the OP approved my answer (at least today 2017-07-13) on the other hand in the comments beneath their question it seems they has of a different opinion.

The answer is however useful in the use case that the sparse matrix represents, e.g., sparse measurements and you want to correct a measurement bias, subtract a mean value, etc. so I'm going to leave it here, even if it can be judged controversial.

Upvotes: 6

Related Questions