search

devsisters / DQN-tensorflow

Tensorflow implementation of Human-Level Control through Deep Reinforcement Learning
MIT License
2.46k stars 765 forks source link

How to generalize to other Atari Games? #20

Closed Chia-Hsuan-Lee closed 7 years ago

Chia-Hsuan-Lee commented 7 years ago

Hi i want to run your code to train agent on other games. Is there any code or hyperparameter that needs to be changed in order to train a nice agent ?

carpedm20 commented 7 years ago

It highly depends on what the game is. If a game needs a long series of steps to make a decision, you need to increase history_length. If the game's state is diverse, you may need to increase memory_size. If the game is hard to learn, you need to increase max_steps. If the input of the game has higher dimension, you need to change the structure of the network to represent the state more clearly.

Chia-Hsuan-Lee commented 7 years ago

Thanks really a lot !