metadata
tags:
- deep-reinforcement-learning
- reinforcement-learning
- stable-baselines3
- atari
- space-invaders
model-index:
- name: dqn-SpaceInvadersNoFrameskip-v4
results:
- task:
type: reinforcement-learning
dataset:
type: SpaceInvadersNoFrameskip-v4
name: Space Invaders (Atari)
metrics:
- type: mean_reward
value: 1175.00 +/ 215.89
name: mean_reward
DQN playing Space Invaders
Trained with Deep Q-Learning (DQN) using stable-baselines3.
Training
- Algorithm: DQN
- Environment: SpaceInvadersNoFrameskip-v4 (Atari)
- Timesteps: 10,000,000
- Environments: 8 parallel
- Batch size: 256
- GPU: Quadro RTX 5000 (16GB)
- Training time: ~5.5 hours
Results
- Final evaluation reward: 1175 +/ 216
- Certification score: 959 (threshold: 200)