DOC: .loc doesn't document that it can take sets or dicts as arguments

[Dr-Irv]

Location of the documentation

https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.loc.html?highlight=loc#pandas.DataFrame.loc

Documentation problem

Surprisingly, the following code works:

import pandas as pd

df = pd.DataFrame([["a", 1], ["b", 2], ["c", 3]], columns=["let", "num"]).set_index(
    "let"
)

dfl = df.loc[set(["a", "b"]), :]

print(dfl)

dd = {"a": 10, "c": 12}

dfd = df.loc[dd, "num"]

print(dfd)

In the documentation, we don't say that you can pass a set or a dict as one of the arguments to .loc .

So either we update the documentation, or maybe we should be raising an Exception?

[attack68]

+1 raising and documenting. also restricting .loc to lists only and not sequence inputs: see #42329 (comment)

[Delengowski]

I'm assuming for speed the implementation of .loc[] tries to iterate rather doing an isinstance check. This is unfortunate because isinstance(dict, typing.Sequence) == False as does isinstance(set, typing.Sequence) iirc.

[phofl]

cc @jbrockmendel Should we decide for 1.4 here?

[jbrockmendel]

it'd be nice to get in, yah

[phofl]

Will look into this the coming days

[nsfinkelstein]

Can I ask what was the reasoning behind deprecating the ability to use a set as an indexer? I've followed the discussion and links above, but cannot discern why the decision was made to deprecate the use of sets as indexers. In my experience, it's incredibly helpful to be able to use sets as indexers, and I'd be happy to contribute code towards un-deprecating this option if possible.

Looking through the code a bit, my understanding is that the index will be converted into a list in any case here: https://github.com/phofl/pandas/blob/f04ec539eaf4ab1d03c1654e6f90b32c6b56f078/pandas/core/frame.py#L3505-L3507.