gyx
gyx copied to clipboard
doctorcorral
Reinforcement Learning environment for Elixir
The DNN handled by a Python process should be training almost all the time. Once a training phase finishes, it immediately update its data batch to train again.
There is a standard way for checking a given action can be processed by the environment. Such action must be an element of the action space defined on the environment....
Tuple spaces can hold a series of [`Discrete`](https://raw.githubusercontent.com/doctorcorral/gyx/master/lib/core/spaces/discrete.ex) and [`Box`](https://raw.githubusercontent.com/doctorcorral/gyx/master/lib/core/spaces/box.ex) spaces. This kind of abstraction is useful to define complex spaces in terms of discrete and box (R^n) sub spaces....
This function is important to check if a proposed action can be performed in an environment.
Describe a replay buffer behavior and implement it in a module for wrapping ETS interactions.
Use [:erlport](http://erlport.org/) to interact with Gym environments on Python instances These Gym calls must be interfaced with `Env` behaviour callbacks
This is a typical example environment that shows fundamental differences between Sarsa and Q-learning strategies. Description of the environment and solution differences are explained in Sutton, Barto (2018), p 132...
Given a set of experiences obtained from step\1 on Env behaviour, update Agent internal Q table representation using `q_set(env_state = %Abstraction{}, action, value)` following constant alpha Monte Carlo update. This...