CI: Unpin NumPy

[TomAugspurger]

Closes #23172

[pep8speaks]

Hello @TomAugspurger! Thanks for submitting the PR.

• There are no PEP8 issues in the file pandas/core/groupby/generic.py !

[jbrockmendel]

Scipy import error on travis

[codecov]

Codecov Report

Merging #23465 into master will increase coverage by 0.02%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #23465      +/-   ##
==========================================
+ Coverage   92.23%   92.25%   +0.02%     
==========================================
  Files         161      161              
  Lines       51197    51176      -21     
==========================================
- Hits        47220    47214       -6     
+ Misses       3977     3962      -15

Flag	Coverage Δ
#multiple	90.64% <100%> (+0.02%)	⬆️
#single	42.28% <33.33%> (+0.01%)	⬆️

Impacted Files	Coverage Δ
pandas/util/_test_decorators.py	93.24% <100%> (ø)	⬆️
pandas/core/groupby/generic.py	87.04% <100%> (ø)	⬆️
pandas/io/parquet.py	83.5% <0%> (+9.77%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0bb24b7...3f84fb0. Read the comment docs.

[TomAugspurger]

I don't really understand the scipy failure. I can't reproduce it locally right now.

I pushed another commit to see if it fails again.

[TomAugspurger]

OK, able to reproduce locally (on a Mac) with the latest wheels from https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com:

$ python -W error
Python 3.7.0 (default, Sep 18 2018, 18:47:22)
[Clang 9.1.0 (clang-902.0.39.2)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import scipy.signal
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/taugspurger/Envs/pandas-dev/lib/python3.7/site-packages/scipy/signal/__init__.py", line 322, in <module>
    from .filter_design import *
  File "/Users/taugspurger/Envs/pandas-dev/lib/python3.7/site-packages/scipy/signal/filter_design.py", line 18, in <module>
    from scipy import special, optimize, fftpack
  File "/Users/taugspurger/Envs/pandas-dev/lib/python3.7/site-packages/scipy/optimize/__init__.py", line 369, in <module>
    from ._minimize import *
  File "/Users/taugspurger/Envs/pandas-dev/lib/python3.7/site-packages/scipy/optimize/_minimize.py", line 30, in <module>
    from ._trustregion_krylov import _minimize_trust_krylov
  File "/Users/taugspurger/Envs/pandas-dev/lib/python3.7/site-packages/scipy/optimize/_trustregion_krylov.py", line 2, in <module>
    from ._trlib import (get_trlib_quadratic_subproblem)
  File "/Users/taugspurger/Envs/pandas-dev/lib/python3.7/site-packages/scipy/optimize/_trlib/__init__.py", line 1, in <module>
    from ._trlib import TRLIBQuadraticSubproblem
  File "messagestream.pxd", line 5, in init scipy.optimize._trlib._trlib
  File "messagestream.pyx", line 5, in init scipy._lib.messagestream
ImportWarning: can't resolve package from __spec__ or __package__, falling back on __name__ and __path__

>>> import numpy
>>> import scipy
>>> numpy.__version__
'1.16.0.dev0+9cc9f01'
>>> scipy.__version__
'1.2.0.dev0+016a6ef'

@rgommers do you know where the configuration for the scipy wheels uploaded to https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com lives? Or do you know offhand what versionsof NumPy and Cython those are compiled with?

[TomAugspurger]

I'm not entirely sure why, but importing here should avoid the test failure. I suspect that whatever modifications pytest makes to the warningsfilter isn't active at this point (in the fixture), so the warning about the dtype size changing isn't elevated to an error.

[rgommers]

@rgommers do you know where the configuration for the scipy wheels uploaded to https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com lives? Or do you know offhand what versionsof NumPy and Cython those are compiled with?

That should be https://github.com/MacPython/scipy-wheels#how-it-works. Haven't checked, but it should be using the lowest supported NumPy version (same as for official releases, for ABI compatibility). Cython version will all be the same, not sure which version (check CI scripts).

[TomAugspurger]

Thanks for the pointer.

From https://travis-ci.org/MacPython/scipy-wheels/jobs/432291485, it seems like Cython 0.26.1 and NumPy 1.14.5 are used.

And FWIW, I get the same warning / error with from python -W error -c "import scipy.signal" using the currently released versions of NumPy & SciPy as I do with the dev versions, so I'll just assume that's expected for now (though I think Cython 0.29 fixed this).

[TomAugspurger]

@jreback if you have any thoughts, otherwise I'll merge tomorrow morning so we're testing against numpy master again.

[jreback]

lgtm.

[TomAugspurger]

In case you didn't follow the issue, the original failure came from us passing generator expressions to vstack, and the new __array_function__ stuff in NumPy dev exhausted those generators. So we just build a list instead (which NumPy was doing anyway)

[jreback]

yep saw that

[jreback]

thanks!