Skip to content

PERF: reducing dtype checking overhead in groupby #44738

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from
Closed

PERF: reducing dtype checking overhead in groupby #44738

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions pandas/core/dtypes/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1223,6 +1223,10 @@ def is_numeric_dtype(arr_or_dtype) -> bool:
>>> is_numeric_dtype(np.array([], dtype=np.timedelta64))
False
"""
try:
return arr_or_dtype.kind in "uifcb"
except AttributeError:
pass
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

comment to the effect of "fastpath/not a dtype object"?

will this change how is_numeric_dtype treats EA dtypes?

return _is_dtype_type(
arr_or_dtype, classes_and_not_datetimelike(np.number, np.bool_)
)
Expand Down
55 changes: 33 additions & 22 deletions pandas/core/groupby/ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@

from pandas._libs import (
NaT,
Period,
lib,
)
import pandas._libs.groupby as libgroupby
Expand All @@ -46,12 +47,10 @@
ensure_int64,
ensure_platform_int,
is_1d_only_ea_obj,
is_bool_dtype,
is_categorical_dtype,
is_complex_dtype,
is_datetime64_any_dtype,
is_float_dtype,
is_integer_dtype,
is_numeric_dtype,
is_sparse,
is_timedelta64_dtype,
Expand Down Expand Up @@ -262,27 +261,35 @@ def _get_output_shape(self, ngroups: int, values: np.ndarray) -> Shape:
out_shape = (ngroups,) + values.shape[1:]
return out_shape

def get_out_dtype(self, dtype: np.dtype) -> np.dtype:
how = self.how

# Note: we make this a classmethod and pass kind+how so that caching
# works at the class level and not the instance level
@classmethod
@functools.lru_cache(maxsize=None)
def get_out_dtype(cls, how: str, dtype: np.dtype) -> np.dtype:
if how == "rank":
out_dtype = "float64"
else:
if is_numeric_dtype(dtype):
if dtype.kind in "uifcb":
out_dtype = f"{dtype.kind}{dtype.itemsize}"
else:
out_dtype = "object"
return np.dtype(out_dtype)

@overload
def _get_result_dtype(self, dtype: np.dtype) -> np.dtype:
@classmethod
def _get_result_dtype(cls, dtype: np.dtype) -> np.dtype:
... # pragma: no cover

@overload
def _get_result_dtype(self, dtype: ExtensionDtype) -> ExtensionDtype:
@classmethod
def _get_result_dtype(cls, dtype: ExtensionDtype) -> ExtensionDtype:
... # pragma: no cover

def _get_result_dtype(self, dtype: DtypeObj) -> DtypeObj:
# Note: we make this a classmethod and pass kind+how so that caching
# works at the class level and not the instance level
@classmethod
@functools.lru_cache(maxsize=None)
def _get_result_dtype(cls, how: str, dtype: DtypeObj) -> DtypeObj:
"""
Get the desired dtype of a result based on the
input dtype and how it was computed.
Expand All @@ -297,8 +304,6 @@ def _get_result_dtype(self, dtype: DtypeObj) -> DtypeObj:
np.dtype or ExtensionDtype
The desired dtype of the result.
"""
how = self.how

if how in ["add", "cumsum", "sum", "prod"]:
if dtype == np.dtype(bool):
return np.dtype(np.int64)
Expand Down Expand Up @@ -382,7 +387,7 @@ def _reconstruct_ea_result(self, values, res_values):
if isinstance(
values.dtype, (BooleanDtype, _IntegerDtype, FloatingDtype, StringDtype)
):
dtype = self._get_result_dtype(values.dtype)
dtype = self._get_result_dtype(self.how, values.dtype)
cls = dtype.construct_array_type()
return cls._from_sequence(res_values, dtype=dtype)

Expand Down Expand Up @@ -422,7 +427,7 @@ def _masked_ea_wrap_cython_operation(
**kwargs,
)

dtype = self._get_result_dtype(orig_values.dtype)
dtype = self._get_result_dtype(self.how, orig_values.dtype)
assert isinstance(dtype, BaseMaskedDtype)
cls = dtype.construct_array_type()

Expand Down Expand Up @@ -490,21 +495,26 @@ def _call_cython_op(
orig_values = values

dtype = values.dtype
is_numeric = is_numeric_dtype(dtype)
dtype_kind = dtype.kind
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

for my own edification, how big a difference does this make? i.e. should i get into the habit of doing this?

# is_numeric_dtype
is_numeric = dtype_kind in "uifcb"

is_datetimelike = needs_i8_conversion(dtype)
# is_datetimelike = needs_i8_conversion(dtype)
is_datetimelike = dtype_kind in ["m", "M"] or (
dtype_kind == "O" and dtype.type is Period
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

at this point we have an ndarray, so PeriodDtype shouldn't be possible i think?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@jorisvandenbossche can you respond to a few small comments here? this should be pretty easy to merge and will be a nice perf bump

)

if is_datetimelike:
values = values.view("int64")
is_numeric = True
elif is_bool_dtype(dtype):
elif dtype_kind == "b":
values = values.astype("int64")
elif is_integer_dtype(dtype):
elif dtype_kind in "ui":
# e.g. uint8 -> uint64, int16 -> int64
dtype_str = dtype.kind + "8"
values = values.astype(dtype_str, copy=False)
elif is_numeric:
if not is_complex_dtype(dtype):
if not dtype_kind == "c":
values = ensure_float64(values)

values = values.T
Expand All @@ -515,7 +525,7 @@ def _call_cython_op(

out_shape = self._get_output_shape(ngroups, values)
func, values = self.get_cython_func_and_vals(values, is_numeric)
out_dtype = self.get_out_dtype(values.dtype)
out_dtype = self.get_out_dtype(self.how, values.dtype)

result = maybe_fill(np.empty(out_shape, dtype=out_dtype))
if self.kind == "aggregate":
Expand Down Expand Up @@ -562,7 +572,7 @@ def _call_cython_op(
# i.e. counts is defined. Locations where count<min_count
# need to have the result set to np.nan, which may require casting,
# see GH#40767
if is_integer_dtype(result.dtype) and not is_datetimelike:
if result.dtype.kind in "ui" and not is_datetimelike:
cutoff = max(1, min_count)
empty_groups = counts < cutoff
if empty_groups.any():
Expand All @@ -575,7 +585,7 @@ def _call_cython_op(
if self.how not in self.cast_blocklist:
# e.g. if we are int64 and need to restore to datetime64/timedelta64
# "rank" is the only member of cast_blocklist we get here
res_dtype = self._get_result_dtype(orig_values.dtype)
res_dtype = self._get_result_dtype(self.how, orig_values.dtype)
op_result = maybe_downcast_to_dtype(result, res_dtype)
else:
op_result = result
Expand Down Expand Up @@ -608,7 +618,8 @@ def cython_operation(
assert axis == 0

dtype = values.dtype
is_numeric = is_numeric_dtype(dtype)
# is_numeric_dtype
is_numeric = dtype.kind in "uifcb"

# can we do this operation with our cython functions
# if not raise NotImplementedError
Expand Down