def _match_join_keys(lcol: ColumnBase, rcol: ColumnBase,
how: str) -> Tuple[ColumnBase, ColumnBase]:
# Casts lcol and rcol to a common dtype for use as join keys. If no casting
# is necessary, they are returned as is.
common_type = None
# cast the keys lcol and rcol to a common dtype
ltype = lcol.dtype
rtype = rcol.dtype
# if either side is categorical, different logic
left_is_categorical = isinstance(ltype, CategoricalDtype)
right_is_categorical = isinstance(rtype, CategoricalDtype)
if left_is_categorical and right_is_categorical:
return _match_categorical_dtypes_both(cast(CategoricalColumn, lcol),
cast(CategoricalColumn, rcol),
how)
elif left_is_categorical or right_is_categorical:
if left_is_categorical:
if how in {"left", "leftsemi", "leftanti"}:
return lcol, rcol.astype(ltype)
common_type = ltype.categories.dtype
else:
common_type = rtype.categories.dtype
return lcol.astype(common_type), rcol.astype(common_type)
if is_dtype_equal(ltype, rtype):
return lcol, rcol
if is_decimal_dtype(ltype) or is_decimal_dtype(rtype):
raise TypeError(
"Decimal columns can only be merged with decimal columns "
"of the same precision and scale")
if (np.issubdtype(ltype, np.number)) and (np.issubdtype(rtype, np.number)):
common_type = (max(ltype, rtype) if ltype.kind == rtype.kind else
np.find_common_type([], (ltype, rtype)))
elif np.issubdtype(ltype, np.datetime64) and np.issubdtype(
rtype, np.datetime64):
common_type = max(ltype, rtype)
if how == "left":
if rcol.fillna(0).can_cast_safely(ltype):
return lcol, rcol.astype(ltype)
else:
warnings.warn(f"Can't safely cast column from {rtype} to {ltype}, "
f"upcasting to {common_type}.")
return lcol.astype(common_type), rcol.astype(common_type)
def fillna(self, fill_value=None, method=None):
if fill_value is not None:
if cudf.utils.utils.isnat(fill_value):
return _fillna_natwise(self)
col = self
if is_scalar(fill_value):
if isinstance(fill_value, np.timedelta64):
dtype = determine_out_dtype(self.dtype, fill_value.dtype)
fill_value = fill_value.astype(dtype)
col = col.astype(dtype)
if not isinstance(fill_value, cudf.Scalar):
fill_value = cudf.Scalar(fill_value, dtype=dtype)
else:
fill_value = column.as_column(fill_value, nan_as_null=False)
return ColumnBase.fillna(col, fill_value)
else:
return super().fillna(method=method)
def _match_join_keys(lcol: ColumnBase, rcol: ColumnBase,
how: str) -> Tuple[ColumnBase, ColumnBase]:
# returns the common dtype that lcol and rcol should be casted to,
# before they can be used as left and right join keys.
# If no casting is necessary, returns None
common_type = None
# cast the keys lcol and rcol to a common dtype
ltype = lcol.dtype
rtype = rcol.dtype
# if either side is categorical, different logic
if isinstance(ltype, CategoricalDtype) or isinstance(
rtype, CategoricalDtype):
return _match_categorical_dtypes(lcol, rcol, how)
if pd.api.types.is_dtype_equal(ltype, rtype):
return lcol, rcol
if isinstance(ltype, cudf.Decimal64Dtype) or isinstance(
rtype, cudf.Decimal64Dtype):
raise TypeError(
"Decimal columns can only be merged with decimal columns "
"of the same precision and scale")
if (np.issubdtype(ltype, np.number)) and (np.issubdtype(rtype, np.number)):
common_type = (max(ltype, rtype) if ltype.kind == rtype.kind else
np.find_common_type([], (ltype, rtype)))
elif np.issubdtype(ltype, np.datetime64) and np.issubdtype(
rtype, np.datetime64):
common_type = max(ltype, rtype)
if how == "left":
if rcol.fillna(0).can_cast_safely(ltype):
return lcol, rcol.astype(ltype)
else:
warnings.warn(f"Can't safely cast column from {rtype} to {ltype}, "
"upcasting to {common_type}.")
return lcol.astype(common_type), rcol.astype(common_type)
