Bandits in the Loop

This module provides tools to design experiments and benchmark multi-armed bandits policies. The idea is to be able to easily evaluate a new algorithm on multiple problems. Either generate a simulated dataset and evaluate online, or load a real dataset and evaluate offline.

The idea is to have the following workflow:

Choose a policy/policies, initialize them

Load dataset or generate synthetic samples

Initialize environment and benchmark parameters (n_runs, results_folder, ...)

Run simulation and get report as plots, comparative HTML&LaTeX table

Project directory structure

environment: defines global parameters, features, feedback graphs etc.
policies: examples policies for different classes of bandits
datasets: fetch, load, generate data
evaluation: setup and run evaluation and comparison loops
utils: helper functions to generate plots and output formatted results (HTML, TeX)

Installation

From `pip`

pip install bitl

From source

Download the source

git clone https://github.com/tritas/bitl.git

Install some dependencies

pip -r install requirements.txt

Install the package

python setup.py install
or
python setup.py develop (--user)

Examples

The examples folder will showcase how to use the different functionality.