The Gym envirment has bug so the training does't give good reward

devsisters / DQN-tensorflow

Tensorflow implementation of Human-Level Control through Deep Reinforcement Learning

MIT License

2.48k stars 763 forks source link

The Gym envirment has bug so the training does't give good reward #36

Open quhezheng opened 7 years ago

quhezheng commented 7 years ago

The best reward would be 30, that all. But by replacing Gym with ROM directly, the output would be very different, very stable reward around 300~400

I dont' know exactly what's wrong with Gym

isVoid commented 7 years ago

Could you explain the steps on replacing Gym with ROM?

hjchen2 commented 7 years ago

I have encountered the same problem that my best reward is only 45 with Gym. If you had find out what's wrong, please let me know. Thanks.

JUZI1 commented 4 years ago

I seem to understand. The new gym environment is different. If you directly use 'Breakout-v0', you will skip four frames in the middle. You should use 'BreakoutNoFrameskip-v0'