README.md

metadata

tags:
  - SpaceInvadersNoFrameskip-v4
  - deep-rl-course
  - dqn
  - reinforcement-learning
  - stable-baselines3
model-index:
  - name: DQN
    results:
      - task:
          type: reinforcement-learning
          name: reinforcement-learning
        dataset:
          name: SpaceInvadersNoFrameskip-v4
          type: SpaceInvadersNoFrameskip-v4
        metrics:
          - type: mean_reward
            value: 340.50 +/- 45.20
            name: mean_reward
            verified: false

DQN Agent playing SpaceInvadersNoFrameskip-v4 👾

This is a trained DQN agent playing SpaceInvadersNoFrameskip-v4. This model was trained as part of the Hugging Face Deep RL Course Unit 3.

Training Details

• Algorithm: Deep Q-Network (DQN)
• Environment: SpaceInvadersNoFrameskip-v4 (Atari)
• Library: Stable Baselines3
• Training timesteps: 1,000,000
• Evaluation: 340.50 +/- 45.20 (10 episodes)

Usage

from stable_baselines3 import DQN
from stable_baselines3.common.env_util import make_atari_env
from stable_baselines3.common.vec_env import VecFrameStack

# Create environment
env = make_atari_env('SpaceInvadersNoFrameskip-v4', n_envs=1)
env = VecFrameStack(env, n_stack=4)

# Load the model
model = DQN.load("dqn-SpaceInvadersNoFrameskip-v4", env=env)

# Enjoy trained agent
obs = env.reset()
for i in range(1000):
    action, _states = model.predict(obs, deterministic=True)
    obs, rewards, dones, info = env.step(action)
    env.render()

This model achieves good performance on the SpaceInvaders Atari game, scoring well above the target score of 200 for the Deep RL Course Unit 3.