keras-gym
keras-gym copied to clipboard
Plug-n-play reinforcement learning with OpenAI Gym and Keras
Implement `ActorCritic` such that the constructor only takes a `FunctionApproximator` (and some extra config parameters). ```python class Func(km.FunctionApproximator): def body(self, S): ... func = Func(lr=0.01) ac = km.ActorCritic(func, gamma=0.9, bootstrap_n=5,...
The probability distributions with differentiable `.sample()` methods are already implemented. It should be just a matter of putting this together. At the user interface level, it would look like: ```python...