meta-BAMDP

Note for readers coming from the arXiv paper:
The version of the code used in the paper is available under the v1.0 release.
For the most up-to-date code, please refer to the main branch.

Project Description and Motivation

In this project we formulate and solve a meta version of a BAMDP problem for two-armed Bernoulli bandit (TABB) task. The file convexity_1comp.py obtains the solution of the meta-BAMDP using the myopic assumption (i.e. k=2 from the manuscript). The file meta_tree.py shows a more general implementation and allows to solve the meta-BAMDP for more relaxed assumptions. All the other files either have helper functions, or are used for making plots.

Prerequisites

To reproduce the data presented in the paper, you need Python 3.11 or above installed and need to install the dependencies specified in requirements.txt

Installation and Usage

Installation

Install the dependencies with

pip install -r requirements.txt

Usage

run

python main.py

to generate all plots found in the paper. This file also shows normalized value obtained as a function of computational cost for the two kinds of approximation schemes mentioned in the manuscript.