Tested using Python version 3.8.3 on Ubuntu 20.04.
First clone the repository, then run
sudo apt-get install libgl1
pip install -r requirements.txt
To run experiment "Test" with 10000 frames on Pong, run the following command:
python code/run_experiments.py --exp-name Test --games Pong --frames 10000
This will produce a folder experiments/<timestamp>_Test
. This folder contains 4 files:
- a .trajectory file, which contains the pickled trajectory to the best cell.
- a .json file, which contains metadata pertaining to the experiment, such as the highscore; the trajectory length; the archive size; the duration; and the number of frames processed.
- a .svg file, which plots the how the algorithm progresses over time.
- a .data file, which contains the pickled raw data used to produce the plots.
Run the help command for more details:
python code/run_experiments.py -h
To watch the agent play out a trajectory, run the following command:
python code/replay_trajectory.py --path <path-to-trajectory.trajectory>
The program will default to code/demo.trajectory
if no argument is given.