Skip to content

Commit

Permalink
PERF-modin-project#7132: Preserve partition lengths in apply_full_axi…
Browse files Browse the repository at this point in the history
…s with keep_partitioning=False (modin-project#7133)

Signed-off-by: Iaroslav Igoshev <iaroslav.igoshev@intel.com>
Co-authored-by: Dmitry Chigarev <dmitry.chigarev@intel.com>
  • Loading branch information
YarShev and dchigarev authored Mar 28, 2024
1 parent 478b86c commit 553eb5c
Showing 1 changed file with 17 additions and 27 deletions.
44 changes: 17 additions & 27 deletions modin/core/dataframe/pandas/dataframe/dataframe.py
Original file line number Diff line number Diff line change
Expand Up @@ -3439,11 +3439,20 @@ def broadcast_apply_full_axis(
index=new_columns,
)
)

if not keep_partitioning:
if kw["row_lengths"] is None and ModinIndex.is_materialized_index(
new_index
is_index_materialized = ModinIndex.is_materialized_index(new_index)
is_columns_materialized = ModinIndex.is_materialized_index(new_columns)
if axis == 0:
if (
is_columns_materialized
and len(new_partitions.shape) > 1
and new_partitions.shape[1] == 1
):
kw["column_widths"] = [len(new_columns)]
elif axis == 1:
if is_index_materialized and new_partitions.shape[0] == 1:
kw["row_lengths"] = [len(new_index)]
if not keep_partitioning:
if kw["row_lengths"] is None and is_index_materialized:
if axis == 0:
kw["row_lengths"] = get_length_list(
axis_len=len(new_index), num_splits=new_partitions.shape[0]
Expand All @@ -3453,11 +3462,7 @@ def broadcast_apply_full_axis(
self._row_lengths_cache
):
kw["row_lengths"] = self._row_lengths_cache
elif len(new_index) == 1 and new_partitions.shape[0] == 1:
kw["row_lengths"] = [1]
if kw["column_widths"] is None and ModinIndex.is_materialized_index(
new_columns
):
if kw["column_widths"] is None and is_columns_materialized:
if axis == 1:
kw["column_widths"] = get_length_list(
axis_len=len(new_columns),
Expand All @@ -3468,42 +3473,27 @@ def broadcast_apply_full_axis(
new_columns
) == sum(self._column_widths_cache):
kw["column_widths"] = self._column_widths_cache
elif len(new_columns) == 1 and new_partitions.shape[1] == 1:
kw["column_widths"] = [1]
else:
if axis == 0:
if (
kw["row_lengths"] is None
and self._row_lengths_cache is not None
and ModinIndex.is_materialized_index(new_index)
and is_index_materialized
and len(new_index) == sum(self._row_lengths_cache)
# to avoid problems that may arise when filtering empty dataframes
and all(r != 0 for r in self._row_lengths_cache)
):
kw["row_lengths"] = self._row_lengths_cache
if (
kw["column_widths"] is None
and ModinIndex.is_materialized_index(new_columns)
and len(new_partitions.shape) > 1
and new_partitions.shape[1] == 1
):
kw["column_widths"] = [len(new_columns)]
if axis == 1:
elif axis == 1:
if (
kw["column_widths"] is None
and self._column_widths_cache is not None
and ModinIndex.is_materialized_index(new_columns)
and is_columns_materialized
and len(new_columns) == sum(self._column_widths_cache)
# to avoid problems that may arise when filtering empty dataframes
and all(w != 0 for w in self._column_widths_cache)
):
kw["column_widths"] = self._column_widths_cache
if (
kw["row_lengths"] is None
and ModinIndex.is_materialized_index(new_index)
and new_partitions.shape[0] == 1
):
kw["row_lengths"] = [len(new_index)]

result = self.__constructor__(
new_partitions, index=new_index, columns=new_columns, **kw
Expand Down

0 comments on commit 553eb5c

Please sign in to comment.