README.md

metadata

library_name: stable-baselines3
tags:
  - Acrobot-v1
  - deep-reinforcement-learning
  - reinforcement-learning
  - stable-baselines3
model-index:
  - name: DQN
    results:
      - metrics:
          - type: mean_reward
            value: '-72.10 +/- 6.44'
            name: mean_reward
        task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: Acrobot-v1
          type: Acrobot-v1

DQN Agent playing Acrobot-v1

This is a trained model of a DQN agent playing Acrobot-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 utils.load_from_hub --algo dqn --env Acrobot-v1 -orga sb3 -f logs/
python enjoy.py --algo dqn --env Acrobot-v1  -f logs/

Training (with the RL Zoo)

python train.py --algo dqn --env Acrobot-v1 -f logs/
# Upload the model and generate video (when possible)
python -m utils.push_to_hub --algo dqn --env Acrobot-v1 -f logs/ -orga sb3

Hyperparameters

OrderedDict([('batch_size', 128),
             ('buffer_size', 50000),
             ('exploration_final_eps', 0.1),
             ('exploration_fraction', 0.12),
             ('gamma', 0.99),
             ('gradient_steps', -1),
             ('learning_rate', 0.00063),
             ('learning_starts', 0),
             ('n_timesteps', 100000.0),
             ('policy', 'MlpPolicy'),
             ('policy_kwargs', 'dict(net_arch=[256, 256])'),
             ('target_update_interval', 250),
             ('train_freq', 4),
             ('normalize', False)])