groupby.py 文件源码

def _do_grouping(self):
        """Group the dataframe
        """
        # First, do groupby on the first key by sorting on the first key.
        # This will sort & shuffle the partitions.
        firstkey = self._by[0]
        df = self._df.sort_value(firstkey)
        groups = df.to_delayed()
        # Second, do groupby internally for each partition.
        @delayed
        def _groupby(df, by):
            grouped = df.groupby(by=by)
            ovdata = _extract_data_to_check_group_overlap(grouped, by)
            return grouped, ovdata

        grouped = [_groupby(g, self._by) for g in groups]
        # Persist the groupby operation to avoid duplicating the work
        grouped = persist(*grouped)
        # Get the groupby objects
        outgroups = list(map(delayed(operator.itemgetter(0)), grouped))
        _check_group_non_overlap_assumption(grouped)
        return outgroups