Multiplying a np.int8 array with 127 yields different numpy array types depending on platform

Question

The following code:

>>> import numpy as np
>>> np.arange(2).astype(np.int8) * 127

produces for numpy 1.13.3

# On Windows
array([0, 127], dtype=int16)
# On Linux
array([0, 127], dtype=int8)

However, if I change the 127 to a 126, both return a np.int8 array. And if I change the 127 to a 128 both return a np.int16 array.

Questions:

Is this expected behaviour?
Why is it different for the two platforms for this one case?

Somehow, numpy.can_cast(127, numpy.int8, casting='safe') returns False on Windows. — user2357112
– user2357112, Commented Dec 6, 2017 at 19:15

user2357112 · Accepted Answer · 2018-02-20 23:33:37Z

18

This is due to NumPy issue 5917. A < instead of <= caused np.can_cast(127, np.int8) to be False, so NumPy used a too-large dtype for 127. The OS-dependence is because C longs have a different size on Linux and Windows, and some NumPy code paths depend on the size of a C long.

A fix has been released in NumPy 1.14.0. Once you update to at least NumPy 1.14.0, you should see a dtype of int8 on all platforms.

edited Feb 20, 2018 at 23:33

answered Dec 6, 2017 at 19:27

user2357112

286k32 gold badges491 silver badges571 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Multiplying a np.int8 array with 127 yields different numpy array types depending on platform

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related