Contextual Bandit

The repo features the experimental analysis on Contextual Combinatorial Cascading Bandits. We highly recommend cite the the paper, using the following bib

@inproceedings{li2016contextual,
  title={Contextual Combinatorial Cascading Bandits},
  author={Li, Shuai and Wang, Baoxiang and Zhang, Shengyu and Chen, Wei},
  booktitle={Proceedings of The 33rd International Conference on Machine Learning},
  pages={1245--1253},
  year={2016}
}

Experiment Dependency

• python3.5
• numpy
• scipy, with community/atlas-lapack-base
• matplotlib, with cairocffi
• colorama

Experiment Data

• Movielens

wget http://files.grouplens.org/datasets/movielens/ml-20m.zip
unzip ml-20m.zip -d movielens

• Rocketfuel

wget http://research.cs.washington.edu/networking/rocketfuel/maps/rocketfuel_maps_cch.tar.gz
wget http://research.cs.washington.edu/networking/rocketfuel/maps/weights-dist.tar.gz
wget http://research.cs.washington.edu/networking/rocketfuel/maps/rocketfuel-traces.tgz
tar -xvf rocketfuel_maps_cch.tar.gz -C isp
tar -xvf weights-dist.tar.gz -C isp-weight
tar -xvf rocketfuel-traces.tgz -C isp-trace

Experiment

• Before running, please modify the code bleedtest.py accordingly, then

python3 bleedtest.py

• Logs in human-understandable format are attached in log