Play with Mab

This project is built to explain the popular Reinforcement Learning framework of Multi Armed Bandit (MAB) with a easy UI.

You can either play the "MAB envinroment" yourself, by playing with the available arms, or simulate one of the available algorithm which are:

• Thompson Sampling
• Upper Confidence Bound 1 (tuned)
• Epsilon-Greedy
• Random Policy

Parameter updates of the algorithm will be shown at each time steps until convergence is reached. Currently there's a fixed time horizon of 100 steps.

Setup

if you want you can setup a python virtual environment: pip -m venv play_with_mab_venv

in case you did you should activate it: . play_with_mab_venv/bin/activate

then

pip install -r requirements.txt

and just run the main

python3 game_core/__main__.py