You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am not sure what's going on - but I see that Pandas just released version 0.20 without (yet) uploading Linux wheels, so the testing is running a source install of Pandas when it usually runs a binary (wheel) install: pandas-dev/pandas#16260
Might be related to this change which used to pass through any NumPy object dtype (which np.dtype(datetime)is), but now only allows it if the original Pandas dtype was also object.
As I also answered on the pandas issue, you can easily fix the test by removing the astype(datetime.datetime).
I think removing it will work just fine, otherwise to have exactly the same you can replace with .astype(object) or .to_pydatetime() (although you might actually want to test with pd.Timestamps, not with datetime.datetime objects, depending on the purpose of the test)
Activity
matthew-brett commentedon May 5, 2017
I am not sure what's going on - but I see that Pandas just released version 0.20 without (yet) uploading Linux wheels, so the testing is running a source install of Pandas when it usually runs a binary (wheel) install: pandas-dev/pandas#16260
dstansby commentedon May 6, 2017
Hmm, still failing and it looks like the build is using a pandas
0.20.1wheel: https://travis-ci.org/matplotlib/matplotlib/jobs/229296183#L885 , but a different error now:QuLogic commentedon May 6, 2017
Might be related to this change which used to pass through any NumPy object dtype (which
np.dtype(datetime)is), but now only allows it if the original Pandas dtype was alsoobject.jorisvandenbossche commentedon May 6, 2017
As I also answered on the pandas issue, you can easily fix the test by removing the
astype(datetime.datetime).I think removing it will work just fine, otherwise to have exactly the same you can replace with
.astype(object)or.to_pydatetime()(although you might actually want to test with pd.Timestamps, not with datetime.datetime objects, depending on the purpose of the test)