| --- |
| library_name: stable-baselines3 |
| pipeline_tag: reinforcement-learning |
| tags: |
| - stable-baselines3 |
| - reinforcement-learning |
| - deep-reinforcement-learning |
| - PPO |
| - LunarLander-v3 |
| --- |
| |
| # PPO agent for LunarLander-v3 |
|
|
| This repository contains a Stable-Baselines3 PPO actor–critic agent trained on `LunarLander-v3`. |
|
|
| ## Evaluation |
|
|
| Deterministic evaluation over 100 fixed-seed episodes: |
|
|
| | Metric | Value | |
| |---|---:| |
| | Mean reward | 280.66 | |
| | Standard deviation | 34.31 | |
| | Course-style score (`mean - std`) | 246.34 | |
| | Episodes scoring at least 200 | 99.0% | |
| | Minimum reward | 4.31 | |
| | Maximum reward | 322.05 | |
|
|
| The candidate was compared with the previous Hub model on the same 100 fixed seeds. The selection metric was `mean_reward` and the observed improvement was +12.575. |
|
|
| ## Architecture |
|
|
| - Algorithm: PPO |
| - Policy: MLP actor–critic |
| - Actor hidden layers: `[128, 128]` |
| - Critic hidden layers: `[128, 128]` |
|
|
| ## Replay |
|
|
| Replay seed: `42` |
| Replay reward: `266.92` |
|
|
| <video controls autoplay loop muted width="640"> |
| <source src="https://huggingface.co/KaptainKris/HuggingFace_RL_Course/resolve/main/replay.mp4" type="video/mp4"> |
| </video> |
|
|
| ## Load the model |
|
|
| ```python |
| from huggingface_hub import hf_hub_download |
| from stable_baselines3 import PPO |
| |
| checkpoint = hf_hub_download( |
| repo_id="KaptainKris/HuggingFace_RL_Course", |
| filename="ppo-LunarLander-v3.zip", |
| ) |
| |
| model = PPO.load(checkpoint) |
| ``` |
