TF-Agents is a library for Reinforcement Learning. I couldn’t find enough documents and articles to define my environment with TF-Agents. So, I leave how to define my environment as my note.
RockScissorsPaperEnv class in the following notebook is my environment. It’s simple Rock scissors paper environment and number of round is 100. And opponent is always Rock.
In my investigation, at least action_spec method (type of actions) , observation_spec method (observable states) , _step method (one step process including reward calculation and determination of termination) and _reset method (reset environment state) are required.