def _get_unsupported_cols(cls, obj):
"""
Return a list of columns with unsupported by OmniSci data types.
Parameters
----------
obj : pandas.DataFrame or pyarrow.Table
Object to inspect on unsupported column types.
Returns
-------
tuple
Arrow representation of `obj` (for future using) and a list of
unsupported columns.
"""
if isinstance(obj, (pandas.Series, pandas.DataFrame)):
# picking first rows from cols with `dtype="object"` to check its actual type,
# in case of homogen columns that saves us unnecessary convertion to arrow table
cols = [
name for name, col in obj.dtypes.items() if col == "object"
]
type_samples = obj.iloc[0][cols]
unsupported_cols = [
name for name, col in type_samples.items()
if not isinstance(col, str)
and not (is_scalar(col) and pandas.isna(col))
]
if len(unsupported_cols) > 0:
return None, unsupported_cols
try:
at = pyarrow.Table.from_pandas(obj)
except (pyarrow.lib.ArrowTypeError, pyarrow.lib.ArrowInvalid) as e:
regex = r"Conversion failed for column ([^\W]*)"
unsupported_cols = []
for msg in e.args:
match = re.findall(regex, msg)
unsupported_cols.extend(match)
if len(unsupported_cols) == 0:
unsupported_cols = obj.columns
return None, unsupported_cols
else:
obj = at
return (
obj,
[
field.name for field in obj.schema
if not isinstance(field.type, pyarrow.DictionaryType)
and field.type.to_pandas_dtype() == np.dtype("O")
and field.type != "string"
],
)
def wrapper(obj1, obj2, *args, **kwargs):
error_str = f"obj1 and obj2 has incorrect types: {type(obj1)} and {type(obj2)}"
assert not (is_scalar(obj1) ^ is_scalar(obj2)), error_str
assert obj1.__module__.split(".")[0] == "modin", error_str
assert obj2.__module__.split(".")[0] == "pandas", error_str
comparator(obj1, obj2, *args, **kwargs)