BUG: `assert_frame_equals` fails when `MultiIndex` elements are `None`

[attack68]

Pandas version checks

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

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

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

df = DataFrame({
    "Currency": ["USD", "USD", "USD"],
    "Collateral": [None, None, None],
    "Payment": [dt(2023, 1, 1), dt(2023, 1, 1), dt(2023, 2, 1)],
    "Cashflow": [100.0, 100.0, 200.0],
})
gf = df.groupby(["Currency", "Collateral", "Payment"], dropna=False)
       .sum()
       .unstack([0, 1])
       .droplevel(0, axis=1)

Issue Description

This regards assert_frame_equal as part of a test suite.

The first problem with regards to testing gf is that I cannot create a MultiIndex and instantiate its dtypes directly.

expected = DataFrame(
    [200.0, 200.0],
    columns=MultiIndex.from_tuples([("USD", np.nan)], names=["Currency", "Collateral"]),
    index=[dt(2023, 1, 1), dt(2023, 2, 1)]
)
assert_frame_equal(gf, expected, atol=1e-5)

>> Traceback..
>       assert_attr_equal("dtype", left, right, obj=obj)
E       AssertionError: MultiIndex level [1] are different
E       
E       Attribute "dtype" are different
E       [left]:  float64
E       [right]: object

But I can manipulate it like this:

expected = DataFrame(
    [200.0, 200.0],
    columns=MultiIndex.from_arrays([
        Index(["USD"], name="Currency"), 
        Index([None], name="Collateral", dtype="float64")
    ]),
    index=[dt(2023, 1, 1), dt(2023, 2, 1)]
)
assert_frame_equal(gf, expected, atol=1e-5)

>> Traceback..
>       assert_frame_equal(result, expected, atol=1e-4)
E       AssertionError: DataFrame.iloc[:, 0] (column name="('USD', nan)") are different
E       
E       Attribute "name" are different
E       [left]:  ('USD', nan)
E       [right]: ('USD', nan)

There is a NA check for name but apparently it doesn't nest within the tuple so that this fails.

Expected Behavior

Should assert True. Na nest should be checked and works fine.

Installed Versions

2.0.3

[phofl]

Is groupby necessary to reproduce? We fixed some bugs in the MultiIndex construction before the 2.1 rc came out

[attack68]

when i next get chance will check on main or rc.

…

On Tue, 15 Aug 2023 at 23:23, Patrick Hoefler ***@***.***> wrote: Is groupby necessary to reproduce? We fixed some bugs in the MultiIndex construction before the 2.1 rc came out — Reply to this email directly, view it on GitHub <#54521 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFZCAKRGRGCMJKXA3IXB6CLXVPSF3ANCNFSM6AAAAAA3OOU2AM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[chaoyihu]

Is groupby necessary to reproduce?

The intention here is to assert equal on two MultiIndex objects of float type. I think the role of groupby here is to circumvent initializing MultiIndex directly with NA values, which will result in an inferred object dtype.

We fixed some bugs in the MultiIndex construction before the 2.1 rc came out

This still exists on main as of 1165859. The assertion failure says:

Attribute "name" are different
[left]:  ('USD', nan)
[right]: ('USD', nan)

because the levels are not the same:

print(gf.columns)
print(gf.columns.levels)
print(gf.columns.codes)
print("===========")
print(expected.columns)
print(expected.columns.levels)
print(expected.columns.codes)
print("===========")
print(gf.columns == expected.columns)
assert_frame_equal(gf, expected)

MultiIndex([('USD', nan)],
           names=['Currency', 'Collateral'])
[['USD'], [nan]]  # here it is a float nan in the level list
[[0], [0]]
===========
MultiIndex([('USD', nan)],
           names=['Currency', 'Collateral'])
[['USD'], []]  # here it is an empty list
[[0], [-1]]
===========
[False]
AssertionError: DataFrame.iloc[:, 0] (column name="('USD', nan)") are different
Attribute "name" are different
[left]:  ('USD', nan)
[right]: ('USD', nan)

The fix should be similar to #59069, by giving NA values a location index of -1 in levels attributes of MultiIndex instance when is it created from groupby.