ImplicitlyNormalizedForecasterWithClipping

In this repository we provide code for experiments with different popular MultiArmBandits solvers and heavy-tailed noise in reward functions. We also demonstrate our novel solver from paper
https://arxiv.org/abs/2305.06743.

In order to run code:

• clone this repo and install requirements.txt

• type command

python3 main.py --n_arms 2 --iterations 8000 --n_runs 100 --plot_num 2 --output output_folder_name

where

• n_arms is number of arms for each bandit
• iterations is number of iterations for each algorithm per one run
• n_runs is number of runs for each algorithm
• plot_num is number of plots that will be saved during experiment
• output is name of output folder for plots