Can you make a PR to completely remove the dependency please?

See also #35709 and #40789

@mihaimaruseac - I did not realize it can be an even greater bug, i.e. an unused dependency... I can only report the mismatch of requirements with the latest version of this dependency, and an easy workaround - replacing the first equality sign in the requirements file with a "greater than" sign:) Are we absolutely sure the dependency is unused?

A few months ago it was unused. I'm quite confident it is unused now too.

A few months ago it was unused. I'm quite confident it is unused now too.

I think some unit tests would fail if the user had no scipy package pre-installed. But you may argue they are to be run only at build time and thus not needed on the typical user machine.... unless she does the build + tests herself.

Some tests where scipy is required:
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/kernel_tests/random/random_binomial_test.py#L53
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/keras/utils/composite_tensor_support_test.py#L386
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/kernel_tests/distributions/student_t_test.py#L46

Yes, scipy is used only at test time. It should not show as install dependency in setup.py though.

@mirekphd
Please update as per above comment.

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

Are you satisfied with the resolution of your issue?
Yes
No

hum, the scipy==1.4.2 dependancy is apparently not removed in Tensorflow_cpu-2.3.0rc2. is it normal ?

The r2.3 branch used for the 2.3 release has

The relaxation only occurred on master.

it may become a problem when pip new strict resolver comes in full force this fall 2020. Is there any hope to remove it for final 2.3.0 ?

Unfortunately we have started the final build steps, so we can only attempt removal on a subsequent patch release.

Can we re-open this issue so it doesn't get lost?

Did it in a better way. Now, if there is a patch release for any of the affected branches the scipy pinning will also be removed.

Did it in a better way. Now, if there is a patch release for any of the affected branches the scipy pinning will also be removed.

Thanks very much for fixing this (really fast too! 🚀), as this is keeping pyhf from using TensorFlow Probability v0.11.0. 👍

I'm seeing similar issue on aarch64 platform.
tensorflow 2.3.0 requires scipy==1.4.1, but you'll have scipy 1.5.3 which is incompatible

Getting error:
Traceback (most recent call last):
File "face_recognition_pymodule_sklearn.py", line 26, in
from sklearn.preprocessing import LabelEncoder
File "/usr/lib/python3.7/site-packages/sklearn/init.py", line 134, in
from .base import clone
File "/usr/lib/python3.7/site-packages/sklearn/base.py", line 11, in
from scipy import sparse
File "/usr/lib/python3.7/site-packages/scipy/init.py", line 151, in
from scipy._lib._ccallback import LowLevelCallable
File "/usr/lib/python3.7/site-packages/scipy/_lib/_ccallback.py", line 1, in
from . import _ccallback_c
ImportError: cannot import name '_ccallback_c' from 'scipy._lib' (/usr/lib/python3.7/site-packages/scipy/_lib/init.py)