Project 2: Lunar Lander

The repo contains code to run experiments required for CS 7642 Summer 2018 Project 2.

• Required packages: Python 3.6.1, Numpy 1.12.1, Pandas 0.20.1, Matplotlib 2.0.2, Gym 0.10.5, Box2D 2.3.2, PyTorch 0.4.0.

• The repo contains these folders:

  • logs: contains csv files that are results from each experiments, the file names are in the format of train_param_<experiment number> or test_param_<experiment number> for training or testing results, where experiment number is a four-digit number corresponding to the row numbers in param_list.csv under params directory. As well as plots generated from these log files
  • params: contains param_list.csv and param_list.json. All hyperparameters used in experiments can be found in these two files, index by experiment number described above.
  • report: full report in PDF and LaTex format, and accessory files
  • weights: saved model and model weights, files named after the experiment number.

• The repo contains these files:

  • README.md: this file, description of this repo
  • learner.py: DQN learner class and necessary facilitating classes, called from run.py. DQN learner class includes both learn() and test() methods, test() is called after learn() is finished
  • tester.py: DQN tester class, used to run additional tests on existing trained model weights.
  • run.py: used to pass hyperparameters to and run DQN learner.
  • plotter.py: plotting functions that are called by other scripts
  • plotlog.py: used to plog a log file under log directory by running python plotlog.py <file name> in terminal, where file name is without extension.

• Perform experiments by running python run.py from terminal, after setting desired hyperparameters in run.py script.