add: another pong example - Githubissues

hunkim / ReinforcementZeroToAll

249 stars 132 forks source link

add: another pong example #17

Closed kkweon closed 7 years ago

kkweon commented 7 years ago

Yet another pong example

Vanilla Pong implementation
Able to beat the AI at around 6 hours training time
I'm still training this one and the a3c simultaneously until I get the maximum score
A3C PR will be next..
As the agent starts to learn how to play, the training time gets longer because the agent and the ai starts playing neck & neck.
Might needs a refactoring

Performance

episode, rewards
2648 3.0
2649 4.0
2650 1.0
2651 8.0
2652 17.0
2653 11.0
2654 7.0
2655 3.0
2656 7.0

Gif

vanilla_pong