Add TaskMetadata and EnsembleScorer to cleanup and streamline the logic in ensemble_selection_config_scorer.py.
The overall logic remains the same, but with a more clean structure. I've confirmed that simulation results remain identical.
Note: GitHub makes this look like a much bigger change than it is. What changed is that the logic in run_task was moved to a dedicated class called EnsembleScorer, which is constructed during init of EnsembleSelectionConfigScorer. TaskMetadata is a convenience wrapper around zeroshot_gt to simplify the code.
Add TaskMetadata and EnsembleScorer to cleanup and streamline the logic in
ensemble_selection_config_scorer.py.The overall logic remains the same, but with a more clean structure. I've confirmed that simulation results remain identical.
Note: GitHub makes this look like a much bigger change than it is. What changed is that the logic in
run_taskwas moved to a dedicated class calledEnsembleScorer, which is constructed during init ofEnsembleSelectionConfigScorer.TaskMetadatais a convenience wrapper aroundzeroshot_gtto simplify the code.