NaT coercion fails when entire column is NaT

[ludaavics]

Describe the bug Schema with coerced, nullable dates choke when the entire column is NaT values

• [x] I have checked that this issue has not already been reported.
• [x] I have confirmed this bug exists on the latest version of pandera.
• [ ] (optional) I have confirmed this bug exists on the master branch of pandera.

Code Sample, a copy-pastable example

import datetime
import pandas as pd
import pandera as pa


class Foo(pa.SchemaModel):

    date: pa.typing.Series[datetime.date] = pa.Field(nullable=True)

    class Config:
        coerce = True

pandas_datetimes = pd.bdate_range("2023-01-01-", "2023-01-31").tolist()
df = pd.DataFrame(pandas_datetimes, columns=["date"])
Foo.validate(df)  # OK
df = pd.DataFrame(pandas_datetimes + [None], columns=["date"])
Foo.validate(df)  # OK
df = pd.DataFrame(pandas_datetimes + [pd.NaT], columns=["date"])
Foo.validate(df)  # OK
df = pd.DataFrame([pd.NaT] * 2, columns=["date"])
Foo.validate(df)  # raises TypeError

Expected behavior

Should just return a columns of NaT's

Desktop (please complete the following information):

• OS: Ubuntu

[cosmicBboy]

hi @ludaavics can you provide the stack trace of the error?

[ludaavics]

TypeError                                 Traceback (most recent call last)
Cell In [16], line 21
     19 Foo.validate(df)  # OK
     20 df = pd.DataFrame([pd.NaT] * 2, columns=["date"])
---> 21 Foo.validate(df)  # raises TypeError

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/model.py:264, in SchemaModel.validate(cls, check_obj, head, tail, sample, random_state, lazy, inplace)
    249 @classmethod
    250 @docstring_substitution(validate_doc=DataFrameSchema.validate.__doc__)
    251 def validate(
   (...)
    259     inplace: bool = False,
    260 ) -> DataFrameBase[TSchemaModel]:
    261     """%(validate_doc)s"""
    262     return cast(
    263         DataFrameBase[TSchemaModel],
--> 264         cls.to_schema().validate(
    265             check_obj, head, tail, sample, random_state, lazy, inplace
    266         ),
    267     )

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/schemas.py:534, in DataFrameSchema.validate(self, check_obj, head, tail, sample, random_state, lazy, inplace)
    521     check_obj = check_obj.map_partitions(
    522         self._validate,
    523         head=head,
   (...)
    529         meta=check_obj,
    530     )
    532     return check_obj.pandera.add_schema(self)
--> 534 return self._validate(
    535     check_obj=check_obj,
    536     head=head,
    537     tail=tail,
    538     sample=sample,
    539     random_state=random_state,
    540     lazy=lazy,
    541     inplace=inplace,
    542 )

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/schemas.py:724, in DataFrameSchema._validate(self, check_obj, head, tail, sample, random_state, lazy, inplace)
    722 for schema_component in schema_components:
    723     try:
--> 724         result = schema_component(
    725             df_to_validate,
    726             lazy=lazy,
    727             # don't make a copy of the data
    728             inplace=True,
    729         )
    730         check_results.append(check_utils.is_table(result))
    731     except errors.SchemaError as err:

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/schemas.py:2138, in SeriesSchemaBase.__call__(self, check_obj, head, tail, sample, random_state, lazy, inplace)
   2127 def __call__(
   2128     self,
   2129     check_obj: Union[pd.DataFrame, pd.Series],
   (...)
   2135     inplace: bool = False,
   2136 ) -> Union[pd.DataFrame, pd.Series]:
   2137     """Alias for ``validate`` method."""
-> 2138     return self.validate(
   2139         check_obj, head, tail, sample, random_state, lazy, inplace
   2140     )

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/schema_components.py:223, in Column.validate(self, check_obj, head, tail, sample, random_state, lazy, inplace)
    219             validate_column(
    220                 check_obj[column_name].iloc[:, [i]], column_name
    221             )
    222     else:
--> 223         validate_column(check_obj, column_name)
    225 return check_obj

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/schema_components.py:196, in Column.validate.<locals>.validate_column(check_obj, column_name)
    195 def validate_column(check_obj, column_name):
--> 196     super(Column, copy(self).set_name(column_name)).validate(
    197         check_obj,
    198         head,
    199         tail,
    200         sample,
    201         random_state,
    202         lazy,
    203         inplace=inplace,
    204     )

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/schemas.py:2057, in SeriesSchemaBase.validate(self, check_obj, head, tail, sample, random_state, lazy, inplace)
   2052     msg = (
   2053         f"expected series '{series.name}' to have type {self._dtype}, "
   2054         + f"got {series.dtype}"
   2055     )
   2056 elif not isinstance(check_output, bool):
-> 2057     _, failure_cases = check_utils.prepare_series_check_output(
   2058         series,
   2059         pd.Series(list(check_output))
   2060         if not isinstance(check_output, pd.Series)
   2061         else check_output,
   2062     )
   2063     failure_cases = reshape_failure_cases(failure_cases)
   2064     msg = (
   2065         f"expected series '{series.name}' to have type {self._dtype}:\n"
   2066         f"failure cases:\n{failure_cases}"
   2067     )

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandera/check_utils.py:113, in prepare_series_check_output(check_obj, check_output, ignore_na, n_failure_cases)
    107 isna = (
    108     check_obj.isna().all(axis="columns")
    109     if isinstance(check_obj, pd.DataFrame)
    110     else check_obj.isna()
    111 )
    112 try:
--> 113     check_output = check_output | isna
    114 except AttributeError:
    115     # convert check_output to numpy for modin compatibility
    116     check_output = check_output.to_numpy() | isna

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/ops/common.py:70, in _unpack_zerodim_and_defer.<locals>.new_method(self, other)
     66             return NotImplemented
     68 other = item_from_zerodim(other)
---> 70 return method(self, other)

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/arraylike.py:78, in OpsMixin.__or__(self, other)
     76 @unpack_zerodim_and_defer("__or__")
     77 def __or__(self, other):
---> 78     return self._logical_method(other, operator.or_)

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/series.py:5634, in Series._logical_method(self, other, op)
   5631 lvalues = self._values
   5632 rvalues = extract_array(other, extract_numpy=True, extract_range=True)
-> 5634 res_values = ops.logical_op(lvalues, rvalues, op)
   5635 return self._construct_result(res_values, name=res_name)

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/ops/array_ops.py:376, in logical_op(left, right, op)
    372 rvalues = right
    374 if should_extension_dispatch(lvalues, rvalues):
    375     # Call the method on lvalues
--> 376     res_values = op(lvalues, rvalues)
    378 else:
    379     if isinstance(rvalues, np.ndarray):

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/arrays/datetimelike.py:1752, in TimelikeOps.__array_ufunc__(self, ufunc, method, *inputs, **kwargs)
   1744 if (
   1745     ufunc in [np.isnan, np.isinf, np.isfinite]
   1746     and len(inputs) == 1
   1747     and inputs[0] is self
   1748 ):
   1749     # numpy 1.18 changed isinf and isnan to not raise on dt64/td64
   1750     return getattr(ufunc, method)(self._ndarray, **kwargs)
-> 1752 return super().__array_ufunc__(ufunc, method, *inputs, **kwargs)

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/arrays/base.py:1629, in ExtensionArray.__array_ufunc__(self, ufunc, method, *inputs, **kwargs)
   1626     if result is not NotImplemented:
   1627         return result
-> 1629 return arraylike.default_array_ufunc(self, ufunc, method, *inputs, **kwargs)

File ~/miniconda3/envs/dev/lib/python3.9/site-packages/pandas/core/arraylike.py:490, in default_array_ufunc(self, ufunc, method, *inputs, **kwargs)
    486     raise NotImplementedError
    488 new_inputs = [x if x is not self else np.asarray(x) for x in inputs]
--> 490 return getattr(ufunc, method)(*new_inputs, **kwargs)

TypeError: ufunc 'bitwise_or' not supported for the input types, and the inputs could not be safely coerced to any supported types according to the casting rule ''safe''

[ludaavics]

@cosmicBboy any thoughts on this ?

[lih7]

This is a bigger issue when we use a Date column with coerce=True, since a column of all invalid dates will get coerced to pd.NaT but then the entire schema fails validation. Here is an example:

`import pandas as pd import pandera as pa import pandera.typing as pt import numpy as np

class DateOutput(pa.SchemaModel): date: pt.Series[pa.dtypes.Date] = pa.Field(nullable=True, coerce=True)

df = pd.DataFrame({"date": ["2021-08-01", np.nan, pd.NaT]})

df_nat = pd.DataFrame({"date": [np.nan, pd.NaT]})

print(DateOutput.validate(df)) # valid with pd.NaT print(DateOutput.validate(df_nat)) # returns pd.NaT not a valid date`

[riziles]

@ludaavics , I ran into this same issue. The problem was using date instead of datetime. This minor fix works:

import datetime
import pandas as pd
import pandera as pa


class Foo(pa.SchemaModel):

    date: pa.typing.Series[datetime.datetime] = pa.Field(nullable=True)

    class Config:
        coerce = True

df = pd.DataFrame([pd.NaT] * 2, columns=["date"])
Foo.validate(df)  # works

[riziles]

@lih7 , pa.dtypes.Date appears to be undocumented. @cosmicBboy , what are your thoughts on adding some sort of warning in the docstring on this class?

[cosmicBboy]

As far as I know pandas doesn't have first class support for datetime.date, and will be represented as an object.

+1 to adding documentation for pa.dtypes.Date. re: the warning, what kind of warning message did you have in mind @riziles ?

[riziles]

In VS Code, if I type from pandera.dtypes import D, then Date is the first autocomplete suggestion. If I hover over it, I can see the docstring Semantic representation of a date data type. Maybe just add Note: not supported by pandas?