FmyzA's picture
Upload PPO agent trained locally (evaluate + replay video + model card)
2d79c90 verified
|
Raw History Blame Contribute Delete
1.84 kB
metadata
library_name: stable-baselines3
tags:
  - InvertedPendulum-v5
  - deep-reinforcement-learning
  - reinforcement-learning
  - stable-baselines3
  - gymnasium
  - ppo
model-index:
  - name: ppo-InvertedPendulum-v5
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: InvertedPendulum-v5
          type: InvertedPendulum-v5
        metrics:
          - type: mean_reward
            value: 1000.00 +/- 0.00
            name: mean_reward
            verified: false

ppo-InvertedPendulum-v5 Agent playing InvertedPendulum-v5

This is a trained model of a PPO agent playing InvertedPendulum-v5 using the stable-baselines3 library and gymnasium.

Evaluation

metric value
mean_reward 1000.00
std_reward 0.00
mean - std 1000.00
episodes 10

Usage (with Stable-Baselines3)

import gymnasium as gym
from stable_baselines3 import PPO
from huggingface_sb3 import load_from_hub

repo_id = "ppo-InvertedPendulum-v5"          # or <your-user>/<repo-name>
filename = "ppo-InvertedPendulum-v5.zip"

checkpoint = load_from_hub(repo_id, filename)
model = PPO.load(checkpoint)
env = gym.make("InvertedPendulum-v5", render_mode="rgb_array")
obs, _ = env.reset()
for _ in range(1000):
    action, _ = model.predict(obs, deterministic=True)
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, _ = env.reset()
env.close()

Files

  • ppo-InvertedPendulum-v5.zip - the trained policy
  • replay.mp4 - a deterministic replay of one episode
  • results.json - raw evaluation numbers
  • config.json - the constructor arguments stored inside the checkpoint