Project 3: RoboCup

The repo contains code to run experiments required for CS 7642 Summer 2018 Project 3.

• Required packages: Python 3.6.5, Numpy 1.15.5, Pandas 0.23.1, Matplotlib 2.2.2, cvxopt 1.2.0, progress 1.4.

  • progress 1.4 can be found here, and source code here.

• The repo contains these folders:

  • results: contains csv files that are results from each experiments, Q tables in the format of .npy, Q-value difference plots, as well as log.csv recording parameters used in all experiments. This folder needs to be created before any experiment can be run.
  • figs: plots used in figures in the report.
  • report: full report in PDF and LaTex format, and accessory files

• The repo contains these files:

  • README.md: this file, description of this repo
  • Q.py: Q learner, with different variants, called from run.py. All learner classes includes both learn() and train() methods.
  • friendQ.py: CE-Q learner, with different variants.
  • foeQ.py: foe-Q learner, with different variants.
  • ceQ.py: CE-Q learner, with different variants.
  • run.py: takes parameters as argument and run experiments
  • run.sh: bash script to run experiments with different parameters in batch.
  • util.py: utility functions, including plotting and logging functions, and linear programming functions for maximin and CE.
  • soccer.py: soccer game environment.

• Perform experiments by running python3 run.py <algo_name> <init_lr> <end_lr> <init_eps> <end_eps> <max_iter> from terminal, or running bash run.sh to perform experiments in batch.