gouxiangchen / dueling-DQN-pytorch

very easy implementation of dueling DQN in pytorch
68 stars 9 forks source link

We called Double DQN (2015) as DDQN. Dueling DQN (2016) be called as DUEL #2

Closed Yonv1943 closed 4 years ago

Yonv1943 commented 4 years ago

Your code is really good. It deserves more stars.
I try your code in other gym environment and it train faster than other Dueling DQN implement.

Besides, I'm confuse with the name of ddqn.py. Then I know that in here, ddqn.py means Dueling DQN instead of Double DQN.

Even in paper "Dueling Network Architectures for Deep Reinforcement Learning". They call Double DQN as "DDQN", and call Dueling DQN as "DUEL". See paper Table. 2.

gouxiangchen commented 4 years ago

Thank you for your suggestion. In fact in my ddqn implementation, Double DQN is also involved. See ddqn.py, 131 ~ 135 lines.

Maybe I should call it dueling DDQN. How about naming it as d3qn.py :joy:

gouxiangchen commented 4 years ago

I realize that authors also used double dqn in the dueling dqn paper. The name "dueling dqn" may include double dqn implicitly. Therefore I rename ddqn.py to dueling_dqn.py in this repo.

Hope it is clear now, thank you very much.

Yonv1943 commented 4 years ago

Thank you for your reply. My question has been solved. I will closed this "Issue".