score_file#

ibl_bwb_eval.scoring.ts1.score_file(task, pred_path, gt_path)[source]#

Score one prediction file against its ground truth file.

Parameters:

task (str) – Flattened task id (e.g. "ts1-choice").
pred_path (str | Path) – Path to a seed_N.safetensors prediction file.
gt_path (str | Path) – Path to the matching ground_truth.safetensors file.

Return type:

dict[str, float]

Returns:

Dict mapping metric name to scalar value.

Raises:

ValueError – If the task declares a mask the ground truth does not carry, or if the mask leaves no timestep to score.