How to avoid summing duplicates when converting COO to CSR matrix

Question

When converting the COO matrix to the CSR format the duplicates get summed. In the earler versions of scipy there was an flag sum_duplicates=False, but now it is gone. Is there a way to convert COO to CSR matrix without summing the duplicates?

Below is an example generating a coo matrix:

from scipy.sparse import csr_matrix,coo_matrix

i_tar = np.array([3,3,3,3,3,6,7,8,8,2,2,2])
j_src = np.array([8,8,8,8,8,5,7,18,18,4,4,4])
data = np.array([3,4,6,8,40,6,12,5,6,2,3,9])

coo_mat = coo_matrix((data, (i_tar,j_src)),shape=(20,20))

print coo_mat

  (3, 8)  3
  (3, 8)  4
  (3, 8)  6
  (3, 8)  8
  (3, 8)  40
  (6, 5)  6
  (7, 7)  12
  (8, 18) 5
  (8, 18) 6
  (2, 4)  2
  (2, 4)  3
  (2, 4)  9

Converting it to the CSR format we find:

print coo_mat.tocsr()

  (2, 4)  14
  (3, 8)  61
  (6, 5)  6
  (7, 7)  12
  (8, 18) 11

which has duplicates summed. I imaging there are good reasons for summing the duplicates, however, it is not what I need.

What is the expected output? The sum is the standard way to show it, Hyuck Kang gives another solution (choose final row). What do you expect? — kabanus

Hyuck Kang Hyuck Kang · Accepted Answer · 2016-10-25T10:40:40

0

votes

Try this code,

coo_mat.todia().tocsr()

How to avoid summing duplicates when converting COO to CSR matrix

1 Answers