An exploratory 2-week project into Reinforcement Learning and the PPO algorithm
-
Updated
Nov 6, 2026 - Python
An exploratory 2-week project into Reinforcement Learning and the PPO algorithm
PPO on Lunar Lander: reward shaping, a reproducible reward-hacking failure, and a fix
A reinforcement learning project that trains DQN and PPO agents to trade SPY using technical indicators and portfolio data. Built with Python, Gymnasium, Stable-Baselines3, PyTorch, yfinance, pandas, and scikit-learn, it compares the agents against traditional trading strategies and evaluates them using financial performance metrics.
A Master Project implementing a Deep Reinforcement Learning (DDPG) agent for transaction-cost-aware option hedging. Features Behavioral Cloning for a "warm start" and is backtested on real-world SPY ETF data.
PPO agent for Lunar Lander showing reward hacking hidden by deterministic evaluation.
Hybrid DDQN + A* agent for autonomous search and rescue in partially observable environments.
An observability-first reinforcement-learning training & benchmarking library, built on PyTorch and Gymnasium.
Standalone Gym-style wrapper for CityFlow traffic signal control.
Vision-based autonomous racing system comparing PPO, DQN, and GAIL with custom reward shaping across CarRacing-v3 and TORCS simulators
Match-3 game environment for developing and evaluating reinforcement learning agents.
To associate your repository with the gymanasium topic, visit your repo's landing page and select "manage topics."