KatherineJG/ppo-LunarLander-v2

PPO agent playing LunarLander-v2

Trained with Stable-Baselines3 for the Hugging Face Deep RL Course, Unit 1.

Evaluation (10 episodes, deterministic actions, separate Monitor-wrapped env)

Run	mean reward	std reward	mean - std
Local evaluation	252.59	22.44	230.16
package_to_hub evaluation	237.24	23.31	213.93

The two runs differ only through episode sampling; 10 episodes is a small sample.

Training

• Algorithm: PPO, MlpPolicy, 1,015,808 timesteps, 16 parallel envs, seed 1
• n_steps=1024, batch_size=64, n_epochs=4, gamma=0.999, gae_lambda=0.98, ent_coef=0.01
• Python 3.12.10, gymnasium 0.29.1, stable-baselines3 2.3.2, huggingface_sb3 3.0, torch 2.15.0, numpy 1.26.4

Usage

import gymnasium as gym
from huggingface_sb3 import load_from_hub
from stable_baselines3 import PPO
from stable_baselines3.common.evaluation import evaluate_policy

path = load_from_hub("KatherineJG/ppo-LunarLander-v2", "ppo-LunarLander-v2.zip")
model = PPO.load(path)
mean, std = evaluate_policy(model, gym.make("LunarLander-v2"), n_eval_episodes=10, deterministic=True)