hipscatalog_gen.selection.reduce_topk_by_group_dask

reduce_topk_by_group_dask(ddf_like, group_col, score_col, order_desc, k_per_group, ra_col, dec_col, tie_col=None)

Keep up to k_per_group rows per group, sorted by score then RA/DEC.

Uses a two-stage exact strategy: 1) per-partition local top-k pruning (no global shuffle), 2) global top-k by group on the pruned collection.

Parameters:

• ddf_like (Any)

• group_col (str)

• score_col (str)

• order_desc (bool)

• k_per_group (Dict[int, int])

• ra_col (str)

• dec_col (str)

• tie_col (str | None)