How urbanintelligence/wpi-ds551-fall26 works
This repository contains reinforcement-learning course projects: tabular dynamic programming, Monte Carlo and temporal-difference learning, and a deep-Q-learning Atari project. The principal workflow is to run or call an algorithm against an environment and obtain a policy, value estimate, or learned agent. The sampled algorithm bodies are largely assignment stubs, so the graph shows documented interfaces and visible orchestration rather than implying completed learning behavior. Project 3 source supports its setup and evaluation wiring; some agent internals and other project READMEs were not sampled.
Main entry points: main.py Run orchestration · agent_dqn.py DQN agent · mdp_dp.py DP algorithms · test.py Episode evaluation
Components 16
Dynamic Programming
- DP algorithms
Project1/mdp_dp.py - Gymnasium MDP
Tabular Learning
- Monte Carlo
Project2/Project2-1/mc.py - Blackjack env
- TD control
Project2/Project2-2/td.py - Cliffwalking env
Atari DQN
- Run orchestration
Project3/main.py - Environment adapter
Project3/environment.py - Atari preprocessing
Project3/atari_wrapper.py - Gymnasium Atari
- DQN agent
Project3/agent_dqn.py - Agent interface
Project3/agent.py - Q network
Project3/dqn_model.py - Episode evaluation
Project3/test.py
Other components
- Course user
- CLI user
Connections 17
- Course user → DP algorithms: calls algorithms
- DP algorithms → Gymnasium MDP: reads dynamics
- DP algorithms → Gymnasium MDP: runs episodes
- Course user → Monte Carlo: calls algorithms
- Monte Carlo → Blackjack env: samples episodes
- Course user → TD control: calls algorithms
- TD control → Cliffwalking env: samples episodes
- CLI user → Run orchestration: starts run
- Run orchestration → Environment adapter: creates
- Environment adapter → Atari preprocessing: builds Atari env
- Atari preprocessing → Gymnasium Atari: wraps and steps
- Run orchestration → DQN agent: creates
- DQN agent → Agent interface: inherits
- DQN agent → Q network: imports model
- Run orchestration → Episode evaluation: evaluates agent
- Episode evaluation → DQN agent: requests actions
- Episode evaluation → Environment adapter: runs episodes