vub-dl / u-cmab

Uplifted Contextual Multi-Armed Bandit
19 stars 4 forks source link

Treatment effect optimisation in dynamic environments

Overview

Code is split in simulation code, found in the folder simulation-code and code for the experiments, found in the folder u-cmab.

To install pylift, we refer to pylift, for cs-um we refer to cs-um

Running the code

All code is provided in Python 3.6.6. Before running any experiments, make sure all dependencies are installed (this could take a few minutes):

pip install -r requirements.txt

and for pylift specifically:

git clone https://github.com/wayfair/pylift
cd pylift
pip install .
cd ..

After installation, all experiments can be run in jupyter notebook:

jupyter notebook

Every figure in the submitted paper corresponds with a notebook, provided at the root of this repository. Note that all notebooks are jupyter notebooks, with the exception of one Wolfram Mathematica notebook (Figure~1.nb). Due to the anonymisation process, notebooks are converted to json. When copying the notebooks, one can save a file as ipynb and open with jupyter notebook. For more information, visit this guide