API/BUG: Index creation with named Index resets name

[sinhrks]

import pandas as pd
idx = pd.Index([1, 2 ,3], name='x')
pd.Index(idx)
# Int64Index([1, 2, 3], dtype='int64')

A point is whether to reset existing Index name in below case. Current impl can't distinguish whether name is not passed, or passed None explicitly.

pd.Index(idx, name=None)
# -> should reset name?

[jreback]

Same for DatetimeIndex. smells like a buggy. Should preserve meta-data in the constructor (for another passed Index)

In [18]: pd.DatetimeIndex(pd.date_range('20130101',periods=3,name='foo'))
Out[18]: DatetimeIndex(['2013-01-01', '2013-01-02', '2013-01-03'], dtype='datetime64[ns]', freq='D')

[mbirdi]

I tried to reproduce the bug and failed. Here is a link the code examples I ran on my laptop.

http://nbviewer.ipython.org/gist/mbirdi/7586ebc290bd90a73ed3

So in my examples I was able to reset the name of the index from 'x' to None.

[jreback]

@mbirdi this reproduces on master

In [58]: idx = pd.Index([1, 2 ,3], name='x')

In [59]: pd.Index(idx)
Out[59]: Int64Index([1, 2, 3], dtype='int64')

In [60]: pd.Index(idx).name

you need another index creation (e.g. with just a pd.Index)

[mbirdi]

@jreback
@sinhrks

Thanks for the feedback.

I see the question up for discussion now. I have never really had to create an index from an index. Is that a common thing to do in pandas? I am trying to picture why I would do that with the datasets I work with and I am coming up blank.

[jreback]

well if I have something that could be a list or an Index then I want to be able to wrap it and preserve meta data. The bottom line is that we want to have API consistency.

[IamGianluca]

I can work on this. I'll send a PR shortly.

[IamGianluca]

@jreback What should be the desired behaviour when the user passes None as the value for the argument name and the existing index we are passing had already a name?

idx = pd.Index([1, 2 ,3], name='x')
pd.Index(idx, name=None)
# -> should reset name?

[shoyer]

Well, one option is to use a sentinel object for the default instead, e.g., _default = object(), and then use def Index(..., name=_object):. Then you can tell the difference between the user passing name=None and not passing it at all. That said, I don't know if this is worth worrying about it. It's probably OK to keep the current name if name=None is passed.

[jreback]

I agree with @shoyer here. The point of this is to preserve the name, so name=None is not passing one (yes could do something like @shoyer suggest, but not worth it IMHO).

[IamGianluca]

Thanks for the clarification @jreback and @shoyer . I will send a pull request this evening.

[sinhrks]

The last question (name=None) is derived from test failures when I tried to fix it by simply retrieving the given Index name. @shoyer 's suggestion should work and be consistent.

[jreback]

closed by #11695