evalution

[majid5776]

hi. what does it mean when in a scenario eval_mean_reward increasing also critic loss increasing? can we call it over-fitting?

[matteobettini]

The evaluation reward should be proportional to the training reward. If it is going up it is great! It means your model is learning well! As far as I know, there is no notion of overfitting in RL as the task you are evaluating and training on is the task you want to solve

[majid5776]

thank you. but after 20 evaluation my agents don't work well on my videos. but as I said my evaluation reward is increasing. does it mean I should increase the num of iterations to converge the evaluation reward?

[matteobettini]

This might be due to how you create your reward functions. There might be conflicting objectives. In general, i cannot provide feedback on custom tasks