DEPR: row-based indexing in DataFrame.__getitem__

[jbrockmendel]

Discussed recently in #9595. The two tests this edits are the only two that go down this path AFAICT.

[jorisvandenbossche]

Let's first discuss this a bit more, as this is a huge change. (I am personally -1 on changing this I think)

[jbrockmendel]

@jorisvandenbossche discussion notwithstanding, can i get your help in troubleshooting the doc build? i put in "okwarnings" to fix the previous set of failures, not clear on why dft_minute is now undefined

[jorisvandenbossche]

I think the whitespace alignment is off here (:okwarning: and dft_minute... should be vertically aligned)

[TomAugspurger]

@jbrockmendel can you give examples of what user-facing behavior is being deprecated? I'd be in favor of deprecating fallback indexing for partial string slicing dataframes with datetime indexes

In [18]: df = pd.DataFrame({"A": [1, 2]}, index=pd.date_range('2000', periods=2))

In [19]: df['2000']
Out[19]:
            A
2000-01-01  1
2000-01-02  2

[TomAugspurger]

Does this cause a warning? It's using loc so I wouldn't think it would.

[jbrockmendel]

i took the kitchen-sink approach to getting the docbuild passing, will figure out which of these can be trimmed away

[jorisvandenbossche]

Repeating what I said above again: please open an issue first to discuss this.

[jreback]

iirc there is an issue about this (though woudl have to search)

[jorisvandenbossche]

The issue Brock mentioned is #9595, but that is much broader (about __getitem in general), so if that is the one you are thinking about, I would propose to create a new, specific issue for this topic

[jreback]

The issue Brock mentioned is #9595, but that is much broader (about __getitem in general), so if that is the one you are thinking about, I would propose to create a new, specific issue for this topic

i believe there is a specific issue this was discussed long ago

[jbrockmendel]

can you give examples of what user-facing behavior is being deprecated? I'd be in favor of deprecating fallback indexing for partial string slicing dataframes with datetime indexes

Exactly the example you gave. The relevant cases are where we have all of the following:

• a DataFrame df
• df.index is either DatetimeIndex or PeriodIndex
• key is a date-like string
• key not in df.columns

(these conditions are basically psuedo-code for convert_to_index_sliceable)

[jbrockmendel]

closing to clear the queue, will reivisit