BUG/TST: added TypeError if object dtypes are detected in dataframe

[sharkipelago]

• closes BUG: round not functioning as expected #55114
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

This PR addresses concern 1 of #55114 - Having consistent behavior with Series.round & DataFrame.round.

My solution was to raise a TypeError in a way similar way to #61206.

I changed the following existing tests, but a little worried that might break some things, so any feedback is appreciated.

1. I deleted tests/frrame/methods/test_round.py's test_round_mixed_type as I felt that test conflicted with the current intended behavior of DataFrame.round
2. I edited tests/copy_view/test_methods.py's test_round as it was using a dataframe with strings and ints in its test

[sharkipelago]

The second concern of the issue #55114 was round() did not work for Series or Dataframe when using decimal.Decimal from python's default decimal module. It seems like this is because both NumPy and pandas give array-like structures of decimal.Decimal objects a dtype of object. When I tested np.array([decimal.Decimal("1.2234242333")]).round(), it raised an error.

If Series.round() and DataFrame.round() not raising an error on decimal.Decimal objects is still wanted, I thought a clean solution would be to make a new custom dtype for decimal.Decimal. However, that seemed like a pretty big change so wanted to check if there was another way I should be thinking about this bugfix.

[jbrockmendel]

i would expect this to attempt to operate pointwise (which would still raise on e.g. strings)

[sharkipelago]

i would expect this to attempt to operate pointwise (which would still raise on e.g. strings)

Do you mean I should rewrite the code so that it attempts to round every column individually and then raise if there is a non-numeric column? As opposed to looking at the self.dtypes.values ?

[sharkipelago]

Ohh because StringDtype also exists? and other non-numeric dtypes outside of object? Could I use pandas.api.types.is_numeric_dtype?

[jbrockmendel]

and other non-numeric dtypes outside of object?

im specifically thinking of object dtype columns containing numeric entries

[sharkipelago]

Ah okay, makes sense.

I think the current behavior for series.round() is to raise when an object dtype column containing numeric entries is called though.

The test_round_dtype_object() test in pandas/tests/series/methods/test_round.py is this i think:

  def test_round_dtype_object(self):
        # GH#61206
        ser = Series([0.2], dtype="object")
        msg = "Expected numeric dtype, got object instead."
        with pytest.raises(TypeError, match=msg):
            ser.round()

Should I submit a PR to change this behavior first before implementing your pointwise solution?