Define public API for internal frame classes

Since we are currently working on Modin documentation (docstrings, rst documentation in docs folder), we need to define public API for the classes that are in modin/engines/base/frame folder in order to expose the right API in rst documentation. I would like to suggest the following changes:

BaseFrameAxisPartition - without changes for now

• apply
• shuffle
• force_materialization
• unwrap

PandasFrameAxisPartition - without changes for now

• apply
• shuffle
• deploy_axis_func
• deploy_func_between_two_axis_partitions

BasePandasFrame

• dtype
• axes
• _apply_index_objs - convert to public
• mask
• from_labels
• to_labels
• reorder_labels
• copy
• combine_dtypes
• astype
• add_prefix
• add_suffix
• _numeric_columns - convert to public
• _fold_reduce - convert to public
• _map_reduce - convert to public
• _map - convert to public
• _fold - convert to public
• filter_full_axis
• _apply_full_axis - convert to public
• _apply_full_axis_select_indices - convert to public
• _apply_select_indices - convert to public
• broadcast_apply
• broadcast_apply_select_indices
• broadcast_apply_full_axis
• _binary_op - convert to public
• _concat - convert to public
• groupby_reduce
• from_pandas
• from_arrow
• to_pandas
• to_numpy
• transpose
• finalize - convert to private or rework so that the partition manager manipulate with the partitions, but not the frame itself does.

BaseFrameManager - convert to BasePandasFrameManager

• preprocess_func
• groupby_reduce
• broadcast_apply_select_indices
• broadcast_apply
• broadcast_axis_partitions
• map_partitions
• lazy_map_partitions
• map_axis_partitions
• simple_shuffle
• concat
• concatenate
• to_pandas
• to_numpy
• from_pandas
• from_arrow
• get_indices
• apply_func_to_select_indices
• apply_func_to_select_indices_along_full_axis
• apply_func_to_indices_both_axis
• binary_operation

BaseFramePartition - convert to BasePandasFramePartition

• get
• apply
• add_to_apply_calls
• drain_call_queue
• wait
• to_pandas
• to_numpy
• mask
• put
• preprocess_func
• length
• width
• empty
• width_extraction_fn - convert to private
• length_extraction_fn - convert to private

cc @devin-petersohn

[devin-petersohn]

@YarShev I suppose that would make sense, there’s no way in those cases to avoid calling it because the metadata needs to be coordinated and rebroadcasted. In that case, I would prefer to change the name to synchronize_labels to be more descriptive (and discourage misuse).

[YarShev]

@devin-petersohn , I saw fit to make the following changes:

• FrameManager -> PartitionManager
• _frame_mgr_cls-> _partition_mgr_cls
• BasePandasFrame -> PandasFrame
• BasePandasFrameManager -> PandasFramePartitionManager
• BasePandasFramePartition -> PandasFramePartition
• DaskFrameManager -> PandasOnDaskFramePartitionManager
• PythonFrameManager -> PandasOnPythonFramePartitionManager
• cuDFOnRayFrameManager -> cuDFOnRayFramePartitionManager
• RayFrameManager -> GenericRayFramePartitionManager
• PandasOnRayFrameManager -> PandasOnRayFramePartitionManager
• PyarrowOnRayFrameManager -> PyarrowOnRayFramePartitionManager
• OmnisciOnRayFrameManager -> OmnisciOnRayFramePartitionManager

Please, look at the changes in the PR.