This is a fork of Nathan Sprague's DQN implementation. It is setup to run benchmarks on various PLE games. All the good stuff is Nathan's, bugs and whatnot are mine!
Please read the text below to install.
This has only been run on Ubuntu 14.04
- GPU
- OpenCV (will replace with scikit.image later)
- Theano
- Lasagne
- PyGame Learning Environment
The script dep_script.sh can be used to install all dependencies under Ubuntu. Should work, havent tested. Use at your own risk.
Use the scripts run_nips.py or run_nature.py to start all the necessary processes:
$ ./run_nips.py -g "Pixelcopter"
$ ./run_nature.py --g "Pixelcopter"
The run_nips.py script uses parameters consistent with the original
NIPS workshop paper. This code should take 2-4 days to complete.
Either script will store output files in a folder prefixed with the
name of the game. Pickled version of the network objects are stored
after every epoch. The file results.csv will contain the testing
output. You can plot the progress by executing plot_results.py:
$ python plot_results.py pixelcopter_05-28-17-09_0p00025_0p99/results.csv
After training completes, you can watch the network play using the
ple_run_watch.py script:
$ python run_watch.py pixelcopter_05-28-17-09_0p00025_0p99/network_file_99.pkl