Added set_names to dask Index.

[matheper]

This PR is related to issue #6315.
It aims to add set_names method to Dask Index.

• Tests added / passed
• Passes black dask / flake8 dask

[TomAugspurger]

Hmm I don't think this is what we want. IIUC, doing

df.index.names = ['a', 'b']

would operate on a new index object, rather than updating df.index. Is that correct?

[matheper]

You are right, it is not covering the user case. I noticed the df.index is a property that always returns a new index, that is why the PR does not support any inplace operation.
I have a new commit that caches the df.index and returns it if it exists, otherwise, it does exactly the same as before.
This way it is possible to have inplace operations, but I am not sure if it will cover all the cases. I'll appreciate any feedback you have. Thanks!

[TomAugspurger]

Hmm I didn't appreciate that you were working via a Series / DataFrame rather than index directly.

In that case, I'd recommend implementing / using DataFrame.rename_axis. Would that suffice for your use case?

[matheper]

I have no specific use case, just tried to replicate what was described in the issue.

From @cgi1 comment, I suppose DataFrame.rename_axis should be enough.

The reason for me, why I actually need to rename the Indexes is, that otherwise the reset_index() fails, because there are two Indexes with the same name (to be merged back from the Index site to the non-Index site - And there can be only one column on the non-Index side per name).

[TomAugspurger]

Thanks. I think that rename_axis will fit better than adding a method to Index, but perhaps @cgi1 can chime in.

[TomAugspurger]

@matheper are you interested in working on this still? I think that rename_axis is the better option.

[jsignell]

I think that we should explicitly not support inplace=True which would make this boil down to just a map_partitions I think. @matheper are you still interested in working on this?

[GPUtester]

Can one of the admins verify this patch?