True. It could also just be fixed... Either by making np.random use
npy_intp internally or by making the randrange type functions switch to 64
bits when the input range is too large.
On May 22, 2015 8:20 AM, "argriffing" notifications@github.com wrote:

According to #4085 #4085 numpy
cannot sample integers larger than 32 bits on systems where long is 32
bits (64 bit Windows but not 64 bit Linux or Mac), causing complications
like scipy/scipy#4552 scipy/scipy#4552 which
attempts to sample the sparsity structure of a large sparse matrix. This
limitation could be documented more clearly on pages like
http://docs.scipy.org/doc/numpy/reference/routines.random.html or
http://docs.scipy.org/doc/numpy/reference/generated/numpy.random.randint.html
.

—
Reply to this email directly or view it on GitHub
#5908.

@njsmith Would this fix be subject to the same constraints as the 'choice' and 'dirichlet' and 'normal' improvements as in #5851 (comment)?

Technically yes, but that doesn't necessarily cause a problem. E.g. letting
randrange accept a larger range won't break any existing streams.
.
(There is also an interesting problem where I bet right now you get
different streams from the same code depending on what platform you're
on... Oops.)
On May 22, 2015 10:12 AM, "argriffing" notifications@github.com wrote:

@njsmith https://github.com/njsmith Would this fix be subject to the
same constraints as the 'choice' and 'dirichlet' and 'normal' improvements
as in #5851 (comment)
#5851 (comment)?

—
Reply to this email directly or view it on GitHub
#5908 (comment).

@njsmith This can be closed.