mrm8488's picture
Initial commit
1046d5c
metadata
library_name: stable-baselines3
tags:
  - EnduroNoFrameskip-v4
  - deep-reinforcement-learning
  - reinforcement-learning
  - stable-baselines3
model-index:
  - name: DQN
    results:
      - metrics:
          - type: mean_reward
            value: 553.80 +/- 125.50
            name: mean_reward
        task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: EnduroNoFrameskip-v4
          type: EnduroNoFrameskip-v4

DQN Agent playing EnduroNoFrameskip-v4

This is a trained model of a DQN agent playing EnduroNoFrameskip-v4 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 EnduroNoFrameskip-v4 -orga mrm8488 -f logs/
python enjoy.py --algo dqn --env EnduroNoFrameskip-v4  -f logs/

Training (with the RL Zoo)

python train.py --algo dqn --env EnduroNoFrameskip-v4 -f logs/
# Upload the model and generate video (when possible)
python -m utils.push_to_hub --algo dqn --env EnduroNoFrameskip-v4 -f logs/ -orga mrm8488

Hyperparameters

OrderedDict([('batch_size', 512),
             ('buffer_size', 100000),
             ('env_wrapper',
              ['stable_baselines3.common.atari_wrappers.AtariWrapper']),
             ('exploration_final_eps', 0.01),
             ('exploration_fraction', 0.1),
             ('frame_stack', 4),
             ('gradient_steps', 1),
             ('learning_rate', 0.0001),
             ('learning_starts', 100000),
             ('n_timesteps', 600000.0),
             ('optimize_memory_usage', False),
             ('policy', 'CnnPolicy'),
             ('target_update_interval', 1000),
             ('train_freq', 4),
             ('normalize', False)])