Trait polars::prelude::PhysicalAggregation

pub trait PhysicalAggregation: Send + Sync {
    fn aggregate(
        &self, 
        df: &DataFrame, 
        groups: &Vec<(u32, Vec<u32, Global>), Global>, 
        state: &ExecutionState
    ) -> Result<Option<Series>, PolarsError>;

    fn evaluate_partitioned(
        &self, 
        df: &DataFrame, 
        groups: &Vec<(u32, Vec<u32, Global>), Global>, 
        state: &ExecutionState
    ) -> Result<Option<Vec<Series, Global>>, PolarsError> { ... }
    fn evaluate_partitioned_final(
        &self, 
        final_df: &DataFrame, 
        groups: &Vec<(u32, Vec<u32, Global>), Global>, 
        state: &ExecutionState
    ) -> Result<Option<Series>, PolarsError> { ... }
}

Required methods

fn aggregate(
    &self,
    df: &DataFrame,
    groups: &Vec<(u32, Vec<u32, Global>), Global>,
    state: &ExecutionState
) -> Result<Option<Series>, PolarsError>

Should be called on the final aggregation node like sum, min, max, etc. When called on a tail, slice, sort, etc. it should return a list-array

Provided methods

fn evaluate_partitioned(
    &self,
    df: &DataFrame,
    groups: &Vec<(u32, Vec<u32, Global>), Global>,
    state: &ExecutionState
) -> Result<Option<Vec<Series, Global>>, PolarsError>

This is called in partitioned aggregation. Partitioned results may differ from aggregation results. For instance, for a mean operation a partitioned result needs to return the sum and the valid_count (length - null count).

A final aggregation can then take the sum of sums and sum of valid_counts to produce a final mean.

fn evaluate_partitioned_final(
    &self,
    final_df: &DataFrame,
    groups: &Vec<(u32, Vec<u32, Global>), Global>,
    state: &ExecutionState
) -> Result<Option<Series>, PolarsError>

Called to merge all the partitioned results in a final aggregate.

Implementors