BUG: Pandas loc can result in ambiguous result not caught by an exception

[komodovaran]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample

Tested on Pandas 1.3.0.

import pandas as pd
import numpy as np
import pytest

sample_name = ["foo", "foo", "foo"]
sample_number = [3, 3, 3]

y_pred = np.array([[0.1, 0.1, 0.1], [0.9, 0.9, 0.9]]).reshape(-1, 2)

# a df with columns '0' and '1'
df = pd.DataFrame(y_pred, index=[sample_name, sample_number])

# Returns a df, as expected
c1 = df.loc["foo", 3]
print(c1)
assert hasattr(c1, "columns")

# Key error, as expected, because there's no entry 2
with pytest.raises(KeyError):
    _c2 = df.loc["foo", 2]

# Now it's a Series of (foo, column 0) instead of a DataFrame!
c3 = df.loc["foo", 0]
print(c3)
assert hasattr(c3, "columns")

Problem description

When converting arrays to DataFrames, the columns will automatically be enumerated. Obviously this problem is somewhat circumventable by not doing all these implicit conversions. However, it doesn't warn you of this, and the case I describe above can return a Series or a DataFrame, depending on which index you query for, which, at least for me, created a decent amount of confusion.

Expected Output

I'd imagine it should raise an ambiguity error, similar to #21080, like

 ValueError: '0' is both an index and a column label.

[phofl]

This is a "feature" I think, but not happy with it either. Is not really ambiguous, but you coud avoid this easily through something like this:

df.loc[("foo", 0), :]

if you would like to select the index or

df.loc[("foo", slice(None)), 0]

if you would like to select the column.

This is much better to read imo.

[rhshadrach]

I think the issue is when a tuple is provided to .loc, is that specifying levels of a MultiIndex or an index-column pair? This appears to be value-dependent behavior to me, not sure I would describe it as a feature.

[phofl]

Based on the ambiguity error Shown above I think this is intended, I don‘t think this is good either