gyx v0.1.1 API Reference

Modules

Fun

This behaviour is intended to be followed for any Environment implementation The most critical function to be exposed is step/1 , which serves as a direct bridge between the environment and any agent

This is data structure for representing an experience piece. This is what is returned to an agent when interacting (calling step/1) with the environment

This module defines the behaviour for a Replay Memory. The role of a Replay Memory is to store experiences coming from one or multiple agents when interacting with their environments. In this way, an agent can sample experiencess directly from a replay memory with different strategies that can improve learning convergence

This Behaivour describes necesary interfaces between environment and agent(s) to be performed when running an experiment

This protocol defines basic functions to interact with action and observation spaces

This space represents a bounded R^n space. These bounds are [0.0, 1.0] by default. Such range can be set with [:low, :high] keys. The shape of such box can be set in :shape :random_algorithm and :seed can be used to set a random key used for reproducibility in sampling the space

This space represents a set of n discrete options. Thus, this space is represented by the number n of available enumerable options. Such options are assumed to be {0,1,...,n-1}

This module contains functions to be shared across all types condiered by all Gyx.Core.Spaces protocols

This space allows to glue together Discrete and Box spaces

This module is an API for accessing Python OpenAI Gym methods

This is an environment implementation of the game of Blackjack as described in Sutton and Barto RL book Example 5.1 cited below

This module implements the FrozenLake-v0 environment according to OpenAI implementation: https://gym.openai.com/envs/FrozenLake-v0/

Implements ReplayMemory behaviour relying on ETS

This module contains auxiliary functions to achieve full compatibility with Gym, including functions to obtain Gyx space representations from Gym space specs

This module contains proxy functions to interact with Python processes via Erlport

This module is intended to be used as a Q table representation. It is based on a single GenServer process, using a Map to hold Q table data as part of process state. Note that this is a hand made version of an Agent OTP implementation, which would be preferable that this

This module describes an entire training process, tune accordingly to your particular environment and agent