Feature/528 allow newest pandas version

[nils-braun]

This builds on top of #562 and fixes the two unit tests.

It also does not require the newest version of pandas, but only to not be 0.24.*

PS: we should really fix the .travis.yaml, I have the strong feeling we are not doing this very optimally... I especially do not like the specification of the versions. Will tackle this later.

@dbarbier Are you fine with this?

[nils-braun]

This should fix #528

[coveralls]

Coverage remained the same at ?% when pulling 12999b4 on feature/528-allow-newest-pandas-version into 11e5c07 on master.

[dbarbier]

@dbarbier Are you fine with this?

@nils-braun Sure, thanks a lot!

[dbarbier]

@nils-braun Failure is caused by a change in pandas 0.25.3, pd.melt now calls np.ravel, and quite surprisingly np.ravel(['index', 1]) returns array(['index', '1'], dtype='<U5'), and pd.melt complains that there is no column named str(1). One might argue that this is a pandas bug, but on the other hand I am not sure whether mixing character and numeric column names is well supported, thus I suggest to use a numeric name for index if columns have numeric names.

[nils-braun]

Yes, thanks for pointing out. I did see it yesterday evening but could not fix it before going to bed.

I will create an issue on the pandas repo nevertheless (pandas-dev/pandas#29718), but I guess you are right: we should fix it also from our side.

[nils-braun]

(hopefully this will all go away once pandas-dev/pandas#28859 is finished)

[Froskekongen]

Would it be possible to do a new release once this PR is merged?

[dbarbier]

Hi @nils-braun do you have any idea why travis builds are failing?

[dbarbier]

Got it, pandas 0.25 requires python >= 3.5.3 https://pandas.pydata.org/pandas-docs/stable/install.html#python-version-support
I can confirm that everything works fine with python 3.5.3.

[Rictus]

Thank you for this work!