kreepy/dqn-LunarLander-v2-hypertuning-v01

DQN Agent playing LunarLander-v2

This is a trained model of a DQN agent playing LunarLander-v2 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

Install the RL Zoo (with SB3 and SB3-Contrib):

pip install rl_zoo3

# Download model and save it into the logs/ folder
python -m rl_zoo3.load_from_hub --algo dqn --env LunarLander-v2 -orga kreepy -f logs/
python -m rl_zoo3.enjoy --algo dqn --env LunarLander-v2  -f logs/

If you installed the RL Zoo3 via pip (pip install rl_zoo3), from anywhere you can do:

python -m rl_zoo3.load_from_hub --algo dqn --env LunarLander-v2 -orga kreepy -f logs/
python -m rl_zoo3.enjoy --algo dqn --env LunarLander-v2  -f logs/

Training (with the RL Zoo)

python -m rl_zoo3.train --algo dqn --env LunarLander-v2 -f logs/
# Upload the model and generate video (when possible)
python -m rl_zoo3.push_to_hub --algo dqn --env LunarLander-v2 -f logs/ -orga kreepy

Hyperparameters

OrderedDict([('batch_size', 9),
             ('buffer_size', 56569),
             ('exploration_final_eps', 0.1),
             ('exploration_fraction', 0.1164397832458963),
             ('exploration_initial_eps', 0.03696153798457299),
             ('gamma', 0.0006190974200887802),
             ('gradient_steps', 9),
             ('learning_rate', 0.011288061590135373),
             ('learning_starts', 15731),
             ('max_grad_norm', 3.705892661777349),
             ('n_timesteps', 10000000.0),
             ('policy', 'MlpPolicy'),
             ('policy_kwargs', 'dict(net_arch=[256, 256])'),
             ('target_update_interval', 218430),
             ('tau', 0.04363931503941886),
             ('train_freq', (9, 'episode')),
             ('normalize', False)])