LSD_bandits

This repository contains the code to reproduce the experiments presented in the paper "A Last Switch Dependent Analysis of Satiation and Seasonality in Bandits" by Laforgue et al. (AISTATS 2022).

This repository contains the following files:

• run_exp.py contains the code to reproduce the experiments presented in Section 4 and Figure 3 of the Supplementary Material.
• run_exp_sup.py contains the code to reproduce the experiment presented in Figure 4 of the Supplementary Material.
• algo_tools.py contains the functions called in the above scripts to implement the algorithms ISI-CombUCB1, CombUCB1, Oracle Greedy, and the two Calibration Sequence approaches.
• data_tools.py contains the functions used to generate the instances of the Bernoulli LSD bandit we test.
• res is the folder where graphs and data are saved.

In order to run the experiment presented in Section 4 of the paper, run the following command (~1min):

$ python run_exp.py

In order to run the same experiment but with the additional benchmark of Calibration Sequence approaches, run the following command (~1.5min):

$ python run_exp.py cs

To run the additional experiment presented in Figure 4 of the Supplementary Material, run the following command (~2min):

$ python run_exp_sup.py

The plots generated are saved in the res folder.