zikangzheng/CartPole-Optuna

🎥 Demo Video

Here is a replay of the trained agent (tuned by Optuna) solving the CartPole-v1 environment.

A2C Agent for CartPole-v1 (Tuned with Optuna)

This is an A2C agent trained on the CartPole-v1 environment. This model was trained using Stable-Baselines3 and the hyperparameters were automatically tuned using Optuna.

• Repository: zikangzheng/CartPole-Optuna
• Environment: CartPole-v1
• RL Algorithm: A2C
• Framework: stable-baselines3
• Tuning Library: Optuna

Installation

pip install stable-baselines3[extra] huggingface_sb3 gymnasium

import gymnasium as gym
from huggingface_sb3 import load_from_hub
from stable_baselines3 import A2C

repo_id = "zikangzheng/CartPole-Optuna"
filename = "a2c_cartpole_optuna_best.zip" 

model = load_from_hub(repo_id, filename)

env = gym.make("CartPole-v1", render_mode="human")
(obs, info) = env.reset()
for _ in range(1000):
    action, _states = model.predict(obs, deterministic=True)
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()

📈 Metric

• Mean Reward: 489.5 +/- 15.2

⚙️ Hyperparameters

{
  "gamma": 0.99,
  "lr": 0.00068,
  "n_steps": 256,
  "max_grad_norm": 0.8,
  "net_arch": "small",
  "activation_fn": "tanh"
}