Question for train.py

[CeyaoZhang]

Why does the validation data directly use compute_metrics_for_classification(y_valid, y_valid_pred) but the test data use submit_predictions_for_test_set(y_test_pred)? It seems that we can also use compute_metrics_for_classification(y_test, y_test_pred).

[guosyjlu]

Yes, you are right. Both are okay. It's just a design to isolate validation and test in case somebody wants to save the prediction for test set.