Gaussian-process Utility Thompson Sampling (GUTS)

This repository contains the code implementing the GUTS algorithm.

Setup

To correctly setup the repository, you need to first download the various git submodules. To do this, you simply have to run the following command in the project's main folder, after cloning the repository:

git submodule update --init --recursive

Requirements

Python 3 is required, to install all dependencies:

pip install -r requirements.txt

Experiments

The main algorithm is located in momabs/guts.py. All parameters used for the bandit settings are listed there.

Simply running the run.sh script will run all the 20-arms experiments, with various cooldown, noise and number of objective settings.

Once the runs have finished, you can use the plot.sh script to generate the diverse plots that compare different settings against each other.

In order to make the 5-arms experiments, small changes need to be made to guts.py:

• uncomment mab.redef_self_5arm_example() (line 161)
• change the cooldown settings from 0,5,10,20 to 0,1,2,3 (line 223)