Reinforcement Learning Framework for PyTorch
Find a file
2020-04-14 16:02:59 -04:00
docs Added network documentation 2020-03-20 20:16:29 -04:00
examples Made Logger global 2020-04-14 15:24:48 -04:00
rltorch Work towards simplifying ReplayMemory 2020-04-14 16:02:59 -04:00
tests Added templates for unit testing and sphinx documentation 2020-03-15 14:27:56 -04:00
.gitignore Added templates for unit testing and sphinx documentation 2020-03-15 14:27:56 -04:00
LICENSE Update LICENSE 2020-03-29 14:10:07 -04:00
Readme.md Made Logger global 2020-04-14 15:24:48 -04:00
setup.py Added templates for unit testing and sphinx documentation 2020-03-15 14:27:56 -04:00
tox.ini Added templates for unit testing and sphinx documentation 2020-03-15 14:27:56 -04:00

rltorch

A reinforcement learning framework with the primary purpose of learning and cleaning up personal scripts.

Installation

From GitHub

pip install git+https://github.com/brandon-rozek/rltorch

Components

Config

This is a dictionary that is shared around the different components. Contains hyperparameters and other configuration values.

Environment

This component needs to support the standard openai functions reset and step.

Network

A network takes a PyTorch nn.Module, PyTorch optimizer, and configuration.

Target Network

Takes in a network and provides methods to sync a copy of the original network.

Action Selector

Typtically takes in a network which it then uses to help make decisions on which actions to take.

For example, the ArgMaxSelector chooses the action that produces the highest entry in the output vector of the network.

Memory

Stores experiences during simulations of the environment. Useful for later training.

Agents

Takes in a network and performs some sort of training upon it.