Edit model card

PPO Agent playing Pendulum-v1

This is a trained model of a PPO agent playing Pendulum-v1 using the stable-baselines3 library and the RL Zoo.

The RL Zoo is a training framework for Stable Baselines3 reinforcement learning agents, with hyperparameter optimization and pre-trained agents included.

Usage (with SB3 RL Zoo)

RL Zoo: https://github.com/DLR-RM/rl-baselines3-zoo
SB3: https://github.com/DLR-RM/stable-baselines3
SB3 Contrib: https://github.com/Stable-Baselines-Team/stable-baselines3-contrib

# Download model and save it into the logs/ folder
python -m rl_zoo3.load_from_hub --algo ppo --env Pendulum-v1 -orga sb3 -f logs/
python enjoy.py --algo ppo --env Pendulum-v1  -f logs/

Training (with the RL Zoo)

python train.py --algo ppo --env Pendulum-v1 -f logs/
# Upload the model and generate video (when possible)
python -m rl_zoo3.push_to_hub --algo ppo --env Pendulum-v1 -f logs/ -orga sb3
from stable_baselines3 import PPO
from stable_baselines3.common.env_util import make_vec_env

# Create the environment
env_id = "Pendulum-v1"
env = make_vec_env(env_id, n_envs=1)

# Instantiate the agent
model = PPO(
    "MlpPolicy",
    env,
    gamma=0.98,
    # Using https://proceedings.mlr.press/v164/raffin22a.html
    use_sde=True,
    sde_sample_freq=4,
    learning_rate=1e-3,
    verbose=1,
)

# Train the agent
model.learn(total_timesteps=int(1e5))

Hyperparameters

OrderedDict([('clip_range', 0.2),
             ('ent_coef', 0.0),
             ('gae_lambda', 0.95),
             ('gamma', 0.9),
             ('learning_rate', 0.001),
             ('n_envs', 4),
             ('n_epochs', 10),
             ('n_steps', 1024),
             ('n_timesteps', 100000.0),
             ('policy', 'MlpPolicy'),
             ('sde_sample_freq', 4),
             ('use_sde', True),
             ('normalize', False)])
Downloads last month
93
Video Preview
loading

Evaluation results