(To appear in: Proceedings of NeurIPS2021)

The code is written in Python 3.7.

You can cite our paper as follows:

@inproceedings{Haddenhorst2021,
  title={Identification of the Generalized Condorcet Winner in Multi-dueling Bandits},
  author={Haddenhorst, Bj{\"o}rn and Bengs, Viktor and H{\"u}llermeier, Eyke},
  booktitle = {Proceedings of Advances in Neural Information Processing Systems 34 (NeurIPS 2021)},
  year={2021},
}

To install requirements:

pip install -r requirements.txt

• (A) To obtain the evaluation results of the algorithms, uncomment the corresponding code in "Neurips2021_experiments.py" and execute it.
• (B) To repeat the empirical comparison of the two lower bounds (Prop. 4.1 and Thm 5.2) for the single bandit case (m=k), simply execute "NeurIPS_LB_comparison.py".

• After repeating all experiments in (A), the results shown in the tables are written saved the following files

• The results for (B) are only shown in the terminal and not saved to any file.

In case of any questions, please contact Björn Haddenhorst (bjoernha@mail.upb.de).