groupby

class pylibcudf.groupby.GroupBy(

Table keys,

null_policy null_handling=null_policy.EXCLUDE,

sorted keys_are_sorted=sorted.NO,

list column_order=None,

list null_precedence=None,

)

Group values by keys and compute various aggregate quantities.

For details, see cudf::groupby::groupby.

Parameters:

keysTable

The columns to group by.

null_handlingnull_policy, optional

Whether or not to include null rows in keys. Default is null_policy.EXCLUDE.

keys_are_sortedsorted, optional

Whether the keys are already sorted. Default is sorted.NO.

column_orderlist[order]

Indicates the order of each column. Default is order.ASCENDING. Ignored if keys_are_sorted is sorted.NO.

null_precedencelist[null_order]

Indicates the ordering of null values in each column. Default is null_order.AFTER. Ignored if keys_are_sorted is sorted.NO.

Methods

aggregate(self, list requests, stream, ...)	Compute aggregations on columns.
get_groups(self, Table values=None, stream, ...)	Get the grouped keys and values labels for each row.
replace_nulls(self, Table value, ...)	Replace nulls in columns.
scan(self, list requests, stream, ...)	Compute scans on columns.
shift(self, Table values, list offset, ...)	Compute shifts on columns.

aggregate(self, list requests: list[GroupByRequest], stream: CudaStreamLike | None = None, DeviceMemoryResource mr=None) → tuple

Compute aggregations on columns.

For details, see cudf::groupby::groupby::aggregate().

Parameters:

requestsList[GroupByRequest]

The list of ~.pylibcudf.groupby.GroupByRequest , each representing a set of aggregations to perform on a given column of values.

streamStream | None

CUDA stream on which to perform the operation.

Returns:

Tuple[Table, List[Table, …]]

A tuple whose first element is the unique keys and whose second element is a table of aggregation results. One table is returned for each aggregation request, with the columns corresponding to the sequence of aggregations in the request.

get_groups(

self,

Table values=None,

stream: CudaStreamLike | None = None,

DeviceMemoryResource mr=None,

) → tuple

Get the grouped keys and values labels for each row.

For details, see cudf::groupby::groupby::get_groups().

Parameters:

valuesTable, optional

The columns to get group labels for. If not specified, None is returned for the group values.

streamStream | None

CUDA stream on which to perform the operation.

Returns:

Tuple[List[int], Table, Table]

A tuple of tables containing three items:

• A list of integer offsets into the group keys/values

• A table of group keys

• A table of group values or None

replace_nulls(self, Table value, list replace_policies: list[ReplacePolicy], stream: CudaStreamLike | None = None, DeviceMemoryResource mr=None) → tuple

Replace nulls in columns.

For details, see cudf::groupby::groupby::replace_nulls().

Parameters:

valuesTable

The columns to replace nulls in.

replace_policiesList[replace_policy]

The policies to use to replace nulls.

streamStream | None

CUDA stream on which to perform the operation.

Returns:

Tuple[Table, Table]

A tuple whose first element is the group’s keys and whose second element is a table of values with nulls replaced.

scan(self, list requests: list[GroupByRequest], stream: CudaStreamLike | None = None, DeviceMemoryResource mr=None) → tuple

Compute scans on columns.

For details, see cudf::groupby::groupby::scan().

Parameters:

requestsList[GroupByRequest]

The list of ~.pylibcudf.groupby.GroupByRequest , each representing a set of aggregations to perform on a given column of values.

streamStream | None

CUDA stream on which to perform the operation.

Returns:

Tuple[Table, List[Table, …]]

A tuple whose first element is the unique keys and whose second element is a table of aggregation results. One table is returned for each aggregation request, with the columns corresponding to the sequence of aggregations in the request.

shift(self, Table values, list offset: list[int], list fill_values: list[Scalar], stream: CudaStreamLike | None = None, DeviceMemoryResource mr=None) → tuple

Compute shifts on columns.

For details, see cudf::groupby::groupby::shift().

Parameters:

valuesTable

The columns to shift.

offsetList[int]

The offsets to shift by.

fill_valuesList[Scalar]

The values to use to fill in missing values.

streamStream | None

CUDA stream on which to perform the operation.

Returns:

Tuple[Table, Table]

A tuple whose first element is the group’s keys and whose second element is a table of shifted values.

class pylibcudf.groupby.GroupByRequest(Column values, list aggregations: list[Aggregation])

A request for a groupby aggregation or scan.

This class is functionally polymorphic and can represent either an aggregation or a scan depending on the algorithm it is used with. For details on the libcudf types it converts to, see cudf::groupby::aggregation_request and cudf::groupby::scan_request.

Parameters:

valuesColumn

The column to aggregate.

aggregationsList[Aggregation]

The list of aggregations to perform.