rename_axis(None) now (0.24) does nothing. Previously it removed the axis name

[tdpetrou]

Code Sample, a copy-pastable example if possible

In 0.23 rename_axis was able to remove the axis name like this:

>>> s = pd.Series([1, 2, 3])
>>> s = s.rename_axis('asdf')
>>> s
asdf
0    1
1    2
2    3
dtype: int64

>>> s.rename_axis(None)
0    1
1    2
2    3
dtype: int64

In 0.24, the mapper parameter is defaulted to None so this does not work. The only way to remove an axis label is to use del s.index.name or s.index.name = None

Problem description

Seems like a regression. I have some code that relies on axis name removal with s.rename_axis(None). Seems like there is no clean way to remove an axis name now.

Expected Output

s.rename_axis(None) should work as it did in 0.23.

[TomAugspurger]

`Series.rename(None)` would be my recommended way of removing the name from a Series. We could use `object()` as the default mapper if we want to support `Series.rename_axis(None)`.

…

On Wed, Jan 30, 2019 at 11:22 AM Ted Petrou ***@***.***> wrote: Code Sample, a copy-pastable example if possible In 0.23 rename_axis was able to remove the axis name like this: >>> s = pd.Series([1, 2, 3])>>> s = s.rename_axis('asdf')>>> s asdf0 11 22 3 dtype: int64 >>> s.rename_axis(None)0 11 22 3 dtype: int64 In 0.24, the mapper parameter is defaulted to None so this does not work. The only way to remove an axis label is to use del s.index.name or s.index.name = None Problem description Seems like a regression. I have some code that relies on axis name removal with s.rename_axis(None). Seems like there is no clean way to remove an axis name now. Expected Output s.rename_axis(None) should work as it did in 0.23. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#25034>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHIke1TnqYETrTFTQkV_LufIcuE2Pwks5vIdTEgaJpZM4aarl4> .

[tdpetrou]

@TomAugspurger That renames the Series name attribute. My example involves renaming the axis name. This would go for both the Series and the DataFrame.

[TomAugspurger]

Ah, sorry misread.

…

On Wed, Jan 30, 2019 at 11:37 AM Ted Petrou ***@***.***> wrote: @TomAugspurger <https://github.com/TomAugspurger> That renames the Series name attribute. My example involves renaming the axis name. This would go for both the Series and the DataFrame. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#25034 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQHItdbUXetlr8P_gUL4rpTDwZuCtoUks5vIdhPgaJpZM4aarl4> .

[jschendel]

The only way to remove an axis label is to use del s.index.name or s.index.name = None

You can also pass lambda x: None for the index argument of rename_axis.

Still should fix the regression of course, but at least this provides method chaining compatible workaround in the meantime:

In [2]: s = pd.Series([1, 2, 3], index=pd.Index(list('abc'), name='foo'))

In [3]: s
Out[3]:
foo
a    1
b    2
c    3
dtype: int64

In [4]: s.rename_axis(index=lambda x: None)
Out[4]:
a    1
b    2
c    3
dtype: int64

[tdpetrou]

@jschendel Thank you for pointing out an alternative. Seems like one way to correct this is to have all the axis level names set equal to None when calling s.rename_axis() or df.rename_axis(). I'm not even sure what semantics would work here.

[jschendel]

Seems like one way to correct this is to have all the axis level names set equal to None when calling s.rename_axis() or df.rename_axis()

I think it could probably be implemented this way, though would require some care in handling all the params (mapper, index, columns, axis) without unnecesarily removing an axis name, e.g. not removing the column name when mapper=None and axis=0.

The PR I opened basically follows @TomAugspurger's suggestion of using a sentinel value, like object(), and using is comparisons, so even if for some crazy reason someone wanted to rename their axis to object() it'd work, as it'd be a different instance and have a different id.

Basically:

In [1]: sentinel = object()

In [2]: def foo(bar=sentinel):
   ...:     if bar is sentinel:
   ...:         return 'sentinel'
   ...:     return bar

In [3]: foo()
Out[3]: 'sentinel'

In [4]: foo(None) is None
Out[4]: True

In [5]: foo(object())
Out[5]: <object at 0x7ff8cde2d8e0>

This implementation has the advantage that df.rename_axis() does nothing, so users have to be explicit about what they want to change, and minimizes the potential for surprising results from a user perspective.