Cannot create sparse matrix from size_t indexes

[david-cortes]

If I try to create a sparse matrix in COOrdinate format with row and column indexes in size_t type, coo_matrix will throw an error as it seems it tries to cast size_t to float64.

The following code illustrates the problem:

import numpy as np
import ctypes
from scipy.sparse import coo_matrix

nrow = 10**4
ncol = 10**5
nnz = 10**6
np.random.seed(1)
row_id = np.random.randint(nrow, size=nnz).astype(ctypes.c_size_t)
col_id = np.random.randint(ncol, size=nnz).astype(ctypes.c_size_t)
vals = np.random.normal(size=nnz)

sp_matrix = coo_matrix((vals, (row_id, col_id)))

Throws:

Traceback (most recent call last):

  File "<ipython-input-1-5bbb7c9f9698>", line 13, in <module>
    sp_matrix = coo_matrix((vals, (row_id, col_id)))

  File "/home/david/anaconda3/lib/python3.6/site-packages/scipy/sparse/coo.py", line 150, in __init__
    self._shape = check_shape((M, N))

  File "/home/david/anaconda3/lib/python3.6/site-packages/scipy/sparse/sputils.py", line 281, in check_shape
    new_shape = tuple(operator.index(arg) for arg in args)

  File "/home/david/anaconda3/lib/python3.6/site-packages/scipy/sparse/sputils.py", line 281, in <genexpr>
    new_shape = tuple(operator.index(arg) for arg in args)

TypeError: 'numpy.float64' object cannot be interpreted as an integer

Adding the shape will however allow to construct this matrix:

coo_matrix((vals, (row_id, col_id)), shape=(nrow, ncol))

<10000x100000 sparse matrix of type '<class 'numpy.float64'>'
	with 1000000 stored elements in COOrdinate format>

The problem is not present when the indexes are of int or long type:

coo_matrix((vals, (row_id.astype(ctypes.c_long), col_id.astype(ctypes.c_long))))

<10000x100000 sparse matrix of type '<class 'numpy.float64'>'
	with 1000000 stored elements in COOrdinate format>

[pv]

The index needs to be signed integer, size_t is unsigned.

The error message is perhaps confusing, but it's because in casting float64 is the next common type uint64 and signed integers can be cast to.

EDIT: ok, the shape check should not fail like this nevertheless