from_ts3

ibl_bwb_eval.scoring.aggregation.from_ts3(raw, task)

Inject task into ts3_scoring.score_dir’s (label, seed) keys.

ts3_scoring.score_dir doesn’t retain task in its raw key (a scoring run only ever covers one TS3 task), so the caller supplies it here. The result is the session-less (label, task, seed) shape aggregate() recognizes and normalizes automatically.

Return type:

dict[tuple[str, str, str, int] | tuple[str, str, int], dict[str, float]]