GPUManager#

The Ray actor-class stores cuDF.DataFrame-s and executes operations on it.

Public API#

class modin.core.execution.ray.implementations.cudf_on_ray.partitioning.gpu_manager.GPUManager(gpu_id)#

Ray actor-class to store cudf.DataFrame-s and execute functions on it.

Parameters:: gpu_id (int) – The identifier of GPU.

apply(first, other, func, **kwargs)#

Apply func to values associated with first/other keys of self.cudf_dataframe_dict with storing of the result.

Store the return value of func (a new cudf.DataFrame) into self.cudf_dataframe_dict.

Parameters:

first (int) – The first key associated with dataframe from self.cudf_dataframe_dict.
other (int or ray.ObjectRef) – The second key associated with dataframe from self.cudf_dataframe_dict. If it isn’t a real key, the func will be applied to the first only.
func (callable) – A function to apply.
**kwargs (dict) – Additional keywords arguments to be passed in func.

Returns:

The new key of the new dataframe stored in self.cudf_dataframe_dict (will be a ray.ObjectRef in outside level).

Return type:

int

apply_non_persistent(first, other, func, **kwargs)#

Apply func to values associated with first/other keys of self.cudf_dataframe_dict.

Parameters:

first (int) – The first key associated with dataframe from self.cudf_dataframe_dict.
other (int) – The second key associated with dataframe from self.cudf_dataframe_dict. If it isn’t a real key, the func will be applied to the first only.
func (callable) – A function to apply.
**kwargs (dict) – Additional keywords arguments to be passed in func.

Returns:

The result of the func (will be a ray.ObjectRef in outside level).

Return type:

The type of return of func

free(key)#

Free the dataFrame and associated key out of self.cudf_dataframe_dict.

Parameters:: key (int) – The key to be deleted.

get_id()#

Get the self.gpu_id from this object.

Returns:: The gpu_id from this object (will be a ray.ObjectRef in outside level).
Return type:: int

get_oid(key)#

Get the value from self.cudf_dataframe_dict by key.

Parameters:: key (int) – The key to get value.
Returns:: Dataframe corresponding to key`(will be a ``ray.ObjectRef` in outside level).
Return type:: cudf.DataFrame

put(pandas_df)#

Convert pandas_df to cudf.DataFrame and put it to self.cudf_dataframe_dict.

Parameters:: pandas_df (pandas.DataFrame/pandas.Series) – A pandas DataFrame/Series to be added.
Returns:: The key associated with added dataframe (will be a ray.ObjectRef in outside level).
Return type:: int

reduce(first, others, func, axis=0, **kwargs)#

Apply func to values associated with first key and others keys of self.cudf_dataframe_dict with storing of the result.

Dataframes associated with others keys will be concatenated to one dataframe.

Store the return value of func (a new cudf.DataFrame) into self.cudf_dataframe_dict.

Parameters:

first (int) – The first key associated with dataframe from self.cudf_dataframe_dict.
others (list of int / list of ray.ObjectRef) – The list of keys associated with dataframe from self.cudf_dataframe_dict.
func (callable) – A function to apply.
axis ({0, 1}, default: 0) – An axis corresponding to a particular row/column of the dataframe.
**kwargs (dict) – Additional keywords arguments to be passed in func.

Returns:

The new key of the new dataframe stored in self.cudf_dataframe_dict (will be a ray.ObjectRef in outside level).

Return type:

int

Notes

If len(others) == 0 func should be able to work with 2nd positional argument with None value.

store_new_df(df)#

Store df in self.cudf_dataframe_dict.

Parameters:: df (cudf.DataFrame) – The cudf.DataFrame to be added.
Returns:: The key associated with added dataframe (will be a ray.ObjectRef in outside level).
Return type:: int