User Guide

Introduction

Welcome to rlberry. Use rlberry’s ExperimentManager to train, evaluate and compare rl agents. Like other popular rl libraries, rlberry also provides basic tools for plotting, multiprocessing and logging . In this user guide, we take you through the core features of rlberry and illustrate them with examples and API documentation .

To run all the examples, you will need to install other libraries like “rlberry-scool” (and others).

The easiest way to do it is :

pip install rlberry[torch,extras]
pip install rlberry-scool

rlberry-scool : It’s the repository used for teaching purposes. These are mainly basic agents and environments, in a version that makes it easier for students to learn.

You can find more details about installation here!

You can find our quick starts here :

• Quick Start for Reinforcement Learning in rlberry
  • Importing required libraries
  • Choosing an RL environment
  • Defining an agent and a baseline
  • Experiment Manager
  • Comparing the expected rewards of the final policies
  • Comparing the agents during the learning period
• Quickstart for Deep Reinforcement Learning in rlberry
  • Imports
  • Reminder of the RL setting
  • Gymnasium Environment
  • Running A2C on Acrobot-v1

Set up an experiment

• How to use an environment
  • Use rlberry environment
  • Use Gymnasium environment
  • Use Atari environment
  • Create your own environment
• How to use an Agent
  • Use rlberry Agent
  • Use StableBaselines3 as rlberry Agent
  • Create your own Agent
  • Use experimentManager
• How to use the ExperimentManager
  • Create your experiment
  • Compare with another agent
  • Output the video
  • Some advanced settings
  • Other information
• How to log your experiment
  • Set rlberry’s logger level
  • Writer
• Visualization of policies and plots of training/evaluation metrics in rlberry
  • Generating videos and gif of the policy of a trained agent
  • Plotting training data and reward curves in rlberry
  • Visualization of evaluations of trained agents in rlberry

Experimenting with Deep agents

(In construction)

Reproducibility

• How to seed your experiment
  • Why use seeding?
  • Basics
  • In rlberry
  • External libraries
• How to save/load an experiment
  • how to save an experiment?
  • How to load a previous experiment?
  • Other information
  • How to save/load an agent only? (advanced users)
• How to export/import data (rlberry data, tensorboard data, …)?
  • How to extract data from the WriterData?
  • How to import data from tensorboard?

Advanced Usage

• Adaptive hypothesis testing for comparison of RL agents with AdaStop
  • Hypothesis testing to compare RL agents
  • Comparison of A2C and PPO from stable-baselines3
  • Result visualisation
• Comparison of Agents
  • Quick reminder on hypothesis testing
  • Multiple agent comparison in rlberry
• How to use the external libraries
  • Using rlberry and Gymnasium
  • Using rlberry and Stable Baselines

• Custom Agents (In construction)

• Custom Environments (In construction)

• Transfer Learning (In construction)

Contributing to rlberry

If you want to contribute to rlberry, check out the contribution guidelines.