Reimplementation of Human-level control through deep reinforcement learning
See our writeup for more details about the project and our implementation.
To launch training, run:
python train.py --replay_buffer_capacity 1e5 --replay_start_size 1e5
To generate figures, run:
python figures.py --n_episodes 100 --n_seconds 45
For each game, the left GIF shows random play, and the right GIF shows the trained agent.
Random Play | Trained Agent |
---|---|
![]() |
![]() |
![]() |
![]() |
- Figure generation sometimes hangs or takes exceedingly long; as a result, we have not yet collated GIFs+scores for all games. After fixing this bug, we can expand our results table and include GIFs
- Training hyperparameters may need further tuning for optimal results
- figure generation should use
pandas.DataFrame.from_dict(...).to_latex()
to produce scores table more easily