This is the code used to generate the experiments and the figures used in our paper "Thompson Sampling For Combinatorial Bandits: Polynomial Regret and Mismatched Sampling Paradox". The paper published in the NeurIPS 2024 Proceedings and is also available at : https://arxiv.org/abs/2410.05441
To generate the experiment run the CTS code notebook in order. Feel free to modifier and play with the parameters. You can also find some experiment on the function
This notebook python implements a minimal version of CTS Beta, BG-CTS, and ESCB for the Two decision problem.
Here are the comparison of the regret dependence on the number of arms for the three algorithms:
