DQN Space Invaders
Trained from scratch in Google Colab with Stable-Baselines3 and Codex assistance for the Hugging Face Deep RL course.
Training timesteps: 100000 Evaluation: 20 deterministic full-game episodes, seed 123, unclipped rewards. Mean reward: 434.75; standard deviation: 118.53770497187804; mean minus standard deviation: 316.21229502812196 See training_script.py for exact settings and evaluation.json for individual measured rewards.
- Downloads last month
- 2
Evaluation results
- mean_reward on SpaceInvadersNoFrameskip-v4self-reported434.75 +/- 118.53770497187804