Vladprime/ppo-LunarLander-v2

PPO LunarLander-v2

Trained from a randomly initialized policy for Vlad Kolomiiets (Vladprime), following Unit 1 of the Hugging Face Deep RL Course. No pretrained policy was used.

Training seed: 20261101. Training environment steps: 1,250,000. Algorithm: Stable-Baselines3 PPO on CPU with eight environments.

Final independent evaluation: 200 fixed seeds from 1000000 through 1000199. Mean reward 263.1597, standard deviation 21.4778, mean minus standard deviation 241.6819. Validation used separate seeds 200000–200039. Full episode returns are saved in results.json.

Reproduce

pip install -r requirements.txt
python train.py