search

TheButlah / makrl

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

A2C Implementation #4

Open TheButlah opened 5 years ago

TheButlah commented 5 years ago

Lets implement that shiz yo

TheButlah commented 5 years ago

Status update: the necessary parallelized and batched environments have been implemented in #25, allowing for synchronous learning methods such as A2C and Synchronous Q Learning. Additionally the model API has been formalized in #23 , enabling Agents to be written that interact with that API. Additionally, a working environment step loop using the BatchedEnv was implemented in #25, with comments for where to place the agent-enviroment interaction. Hence all that is left for any synchronous RL method, either this one or #6 is the actual Agent API and implementation, as well as a suitable concrete Model.