Experiments of Reinforcement Learning in Rust

Experiments to learn about Reinforcement Learning

About

Four environments were implemented:

The environment's implementation was based on the Gymnasium library for Python.

About the features implemented:

Only tabular policies: Basic Policy and Double Policy (for double learning)
Two action selections strategies: ε-greedy with uniform distribuition and Upper Confidence Bound (UCB).
Four policy update strategies: One-Step Sarsa, One-Step QLearning, One-Step Expected Sarsa, Sarsa with Eligibility Traces and QLearning with Eligibility Traces.
Three charts for visualizations about the agent's training: Episode's length, Rewards and Training error.
Visualizations for the states of each environment.
A small CLI program to change the parameters used on the training and generate the charts.

First install Rust:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

Then compile the source code:

cargo build -r

To run the experiments use:

./target/release/reinforcement_learning 0

Where '0' is the identifier for the BlackJack env, to see the other options use:

./target/release/reinforcement_learning --help