Reinforcement Learning Algorithms

This repository contains implementations of various reinforcement learning algorithms. Below are the instructions for training and evaluating each algorithm.

WandB Report

For a detailed report on the performance of the algorithms, please refer to the WandB report.

Prerequisites

• UV

Installation

git clone git@github.com:Colinster327/maze-rl.git
cd maze-rl
uv sync
source .venv/bin/activate

PPO (Proximal Policy Optimization)

• Train PPO:

  make ppo_train

• Evaluate PPO:

  make ppo_eval

• 

A2C (Advantage Actor-Critic)

• Train A2C:

  make a2c_train

• Evaluate A2C:

  make a2c_eval

• 

DT (Decision Transformer)

• Collect Data for DT:

  • Note: Ensure the PPO model is trained before collecting data.

  make dt_collect

• Train DT:

  make dt_train

• Evaluate DT:

  make dt_eval

• 

Q-Learning

• Train Q-Learning:

  make qlearn_train

• Evaluate Q-Learning:

  make qlearn_eval

•