I am training mario game in retro using ppo2 baselines for some time. I have tried level3 and level1 too. But even after full training when I play using saved checkpoints, the mario is not able to finish the level. Mostly it falls into the hole or does not even reach half of first easy level.
Do I need to write reward scripts? Or tune some more hyperparameters?
If someone has experience using gym, retro or training such games, then they can suggest.
The code for training:
python -m baselines.run --alg=ppo2 --env=SuperMarioBros-Nes --gamestate=Level1-1.state --num_timesteps=1e7
The code for running mario using saved checkpoint:
python -m baselines.run --alg=ppo2 --env=SuperMarioBros-Nes --gamestate=Level1-1.state --num_timesteps=1e7 --load_path=C:\Users\SANDEE~1.BHU\AppData\Local\Temp\openai-2019-05-09-18-26-28-156556\checkpoints\08300 --play --num_env=1