Experiments

How to reproduce experimental results of Stochastic bandits with groups of similar arms submitted paper ?

Section 5 of the paper

To reproduce all of the empirical results (and more) of the section 5 (Experiments) of the submitted paper, it is enough to run the regret_experiment.ipynb notebook.

Appendix D of the paper

To reproduce all of the empirical results (and more) of the appendix D (Experiments) of the submitted paper, you may run the regret_experiment.ipynb notebook as well as the dispatching_experiment.ipynb notebook.

How to?

To run the jupyter notebooks, you may use the jupyter software (https://jupyter.org/) or use an online plateform such as google Colab (https://colab.research.google.com/)

Which versions

The code has run on google Colab on Friday 4th, June without problem. The stability of the package that were used (numpy, matplotlib, itertools (built-in), datetime (built-in)) makes it very likely to run correctly before the next major updates of one of those packages.

Otherwise, the code has been tested locally on a machine with the following versions : python 3.9.5 numpy 1.20.3 matplotlib 3.4.1

Where to look at?

The algorithms.py file contains all the algorithms that were used in the experiments of this paper. All the sequential algorithms depends on SequentiAlg, a class that is defined in the Forban module.

In the Forban module, a class for Bandit configurations is also defined.