In the custom_loss_model metric.py, what is the best and direct way so that the recorded metric can be shown in tensorboard, without calculating the metrics again like custom_metrics and callbacks example?
I cannot find how to find the metrics which are already computed by the custom model.
def metrics(self):
return {
"policy_loss": self.policy_loss_metric,
"imitation_loss": self.imitation_loss_metric,
}
Thanks in advance.
Hey @Shanchao_Yang , thanks for the question and sorry for the delay. Could you make sure that for future posts you add a category to it (e.g. “RLlib”)? That way, we’ll find it better and can answer more quickly. Thanks
I think this could possibly be a bug on RLlib. Model metrics should automatically show up in TB. …
Thanks. And sorry for the lack of category of Rllib. I would be more careful in future.
Back to the question, yeah, I indeed remember that the imitation loss metric (Pytorch version) is not automatically included in TB…