Scikit-learn 0.21.1 strange failure using nosetests with import train_test_split #13943
Description
Description
Hey there, I run a simple test suite using nose over my package gplearn and the CRON scheduled tests came up with a failure today. I narrowed the test scopes and removed the test that used that function in a dummy PR on my package and it appears simply importing test_train_split in the test code causes a failure. This only occurs with nosetests, pytest shows no failures locally, while pytest passes cleanly locally. My local configuration is linux, built from source. My travis config is all on pre-built conda packages.
This issue does not occur when running things as code, only through nosetest that I can tell. Tests using test_train_split pass just fine, but then a new "test" appears automagically that fails right after the test file has completed all other tests in the file.
The reason this is happening is a bit of a mystery to me, I don't think I'm doing anything unusual in my test configuration. #13483 appears to be the only recent PR to touch this code. I don't see any reason why that should produce these results though.
See here for the failures which only occur on 0.21.1: https://travis-ci.org/trevorstephens/gplearn/builds/537063072 which is running trevorstephens/gplearn#167
Steps/Code to Reproduce
from sklearn.model_selection import train_test_split
def test_unrelated_fail():
return None
nosetest -v -s test.py
Expected Results
No error is thrown due to simply importing test_train_split
Actual Results
======================================================================
ERROR: Split arrays or matrices into random train and test subsets
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/trev/.virtualenvs/ve3/lib/python3.6/site-packages/nose/case.py", line 198, in runTest
self.test(*self.arg)
File "/home/trev/.virtualenvs/ve3/lib/python3.6/site-packages/nose/util.py", line 620, in newfunc
return func(*arg, **kw)
File "/bigdrive/git/scikit-learn/sklearn/model_selection/_split.py", line 2086, in train_test_split
raise ValueError("At least one array required as input")
ValueError: At least one array required as input
----------------------------------------------------------------------
Versions
Seems to only affect 0.21.1 on both python 3.6, 3.7