/onoff_experiments

Experiments for the online and offline CappedIGW algorithm.

Primary LanguagePython

Infinite Action Bandits with Data Reuse

This repo contains two example projects of the CappedIGW algorithm.

The first project is in the demo directory. This project was made specifically for new users and contains:

  1. Two demonstration notebooks that make it easy to play with CappedIGW on real world datasets
  2. Simplified and documented implementations of CappedIGW and Betting Martingale Normalization.

The second project is in the paper directory. This project contains code to reproduce the results in the published paper. This code is harder to read than the implementations in the demo directory and also includes a working implementation of the algorithm to adaptively choose $\tau$.

Getting Started With The Demo Project:

To play with the experiments in the demo directory follow these steps:

  1. Download this repo to your local machine
  2. Make sure you have python installed
  3. On the command line run pip install notebook
  4. On the command line navigate to your download of the repo
  5. On the command line run jupyter notebook
  6. From your web browser open either of the notebook files in demo

Getting Started With The Paper Project:

To run the experiments in the paper directory follow these steps:

  1. Download this repo to your local machine
  2. Make sure you have python installed
  3. On the command line navigate to your download of the repo
  4. On the command line run pip -r ./paper/requirements.txt
  5. On the command line run python ./paper/run_online.py
  6. On the command line run python ./paper/run_offline.py
  7. Open plots.ipynb to create plots from the paper