search

TheButlah / makrl

makrl - modular algorithm kit for reinforcement learning
4 stars 1 forks source link

DQN Implementation #8

Closed bayoumi17m closed 5 years ago

bayoumi17m commented 5 years ago

I implemented the DQN, along with it, there are boolean conditions in the class that allow us to very easily switch between the Natural DQN, Dueling DQN, DQN with Prioritized Memory, and Dueling DQN with Prioritized Memory. The train_test script runs all the options on MsPacman-ram-v0 and plots the cost and accumulated reward.

TheButlah commented 5 years ago

You should commit to DQN, you don't need to make pull requests for that, only for dev and master