ForcesTrainer error

[kz29]

I am getting the following error when I use ForcesTrainer on different models. I am trying to run train_s2ef_example from the tutorial folder. Any help will be appreciated.

427 if self.scheduler.scheduler_type == "ReduceLROnPlateau":
428 if self.step % eval_every == 0:
429 self.scheduler.step(
--> 430 metrics=val_metrics[primary_metric]["metric"],
431 )
432 else:
433 self.scheduler.step()

UnboundLocalError: local variable 'val_metrics' referenced before assignment

[abhshkdz]

Hey, have you specified a validation dataset? ReduceLROnPlateau uses val metrics to decide when to step the learning rate. Not specifying a validation set would raise the error you're seeing.

Alternatively, if you want to train without a validation set, try setting 'scheduler': "Null", in the optimizer dictionary. That'd use the same learning rate without any schedule on it.

[kz29]

Thank you for the explanation. It is working now.