BUG: to_numeric fails to convert a Pyarrow Decimal series containing NA values #61659
Conversation
| if is_numeric_dtype(values_dtype): | ||
| if is_series: | ||
| return arg._constructor(values, index=arg.index, name=arg.name) | ||
| else: | ||
| return values |
There was a problem hiding this comment.
Could you try to specify if not is_numeric_dtype here, and let the logic below try to convert this case anyways? This short circuit would skip some of the options set in to_numeric
There was a problem hiding this comment.
also as @mroeschke suggested in #61641 (comment) documentation update would also be welcome
There was a problem hiding this comment.
Thanks @simonjayhawkins , I already create anothter PR for document issue: #61904
simonjayhawkins
added
Bug
Dtype Conversions
| result = to_numeric(series, errors="coerce") | ||
|
|
||
| expected = Series([1.00, pd.NA], dtype=decimal_type) | ||
| tm.assert_series_equal(result, expected) |
There was a problem hiding this comment.
IIUC we want a no-op here?
so does
| tm.assert_series_equal(result, expected) | |
| tm.assert_series_equal(result, series) |
work?
| def test_to_numeric_arrow_decimal_with_na(): | ||
| # GH 61641 | ||
| pa = pytest.importorskip("pyarrow") | ||
| decimal_type = ArrowDtype(pa.decimal128(3, scale=2)) |
There was a problem hiding this comment.
maybe we could add the dtype as parameterization and also add any other EA dtypes that are expected to no-op?
so we could call the test something like.. test_to_numeric_EA_is_numeric?
| if is_numeric_dtype(values_dtype): | ||
| if is_series: | ||
| return arg._constructor(values, index=arg.index, name=arg.name) | ||
| else: | ||
| return values |
There was a problem hiding this comment.
also as @mroeschke suggested in #61641 (comment) documentation update would also be welcome
to_numericfails to convert a Pyarrow Decimal series containing NA values. #61641doc/source/whatsnew/v3.0.0.rstfile if fixing a bug or adding a new feature.