xtma / pytorch_car_caring Goto Github PK

Hi, I'm new to machine learning and I'm trying to solve the car-racingv0 environment using PPO.
I've tested this solution and in a matter of minutes my RAM consumptions reaches 16GB and the training stops. Is it normal? Does anyone know how to solve this?

Moving average score is defined as running_score.
running_score = running_score * 0.99 + score * 0.01
The score is the evaluation of the agent's performance in the current episode.
I think the current score is more important than those in the history.
So, why is the weight of running score higher than the weight of score？

Hi,

I notice that you've used some reward shaping during training, but forgot to remove it in in calculating the running average and in test.py, and that leads to incorrect results.
Commenting out relevant lines, I got an average score of around 860 over 100 episodes for the model in the repo.