Report Tune metrics in final evaluation (#7507)

2020-10-01 09:52:36 -04:00
parent 9a92afb6d0
commit dabc85d1ba
1 changed files with 1 additions and 2 deletions
@@ -124,8 +124,7 @@ def run_hp_search_ray(trainer, n_trials: int, direction: str, **kwargs) -> BestR
            metrics = trainer.evaluate()
            trainer.objective = trainer.compute_objective(metrics)
            trainer._tune_save_checkpoint()
-            ray.tune.report(objective=trainer.objective)
-        return trainer.objective
+            ray.tune.report(objective=trainer.objective, **metrics, done=True)

    # The model and TensorBoard writer do not pickle so we have to remove them (if they exists)
    # while doing the ray hp search.