Conciliator steering

The project is a code implementation for the TCML paper "Conciliator steering: Imposing user preference in MORL decision-making problems". The code enables the user to interactively or pre-definedly impose a priority weighting over rewards in a DeepSeaTreasure v1 benchmark presented by Cassimon et al., producing policies resulting in the user's preferred outcome.

This paper is an output of the AIforLEssAuto project. If you wish to see other outputs of the project, check out this repository.

Getting started

Prerequisites

• For Python, follow the insctructions from Python's official website.

• For Git on Windows, follow the instructions from Git's official website.

Note: the code is designed in a Windows environment using the PyPI package installer due to the pip install commands included in the command line script. Using this script in other environments may induce errors.

Usage

1. Make sure the prerequisites are installed
2. Clone or download the repository
3. Run sh commands.sh in the command line in the root directory
4. Success!

Structure

The repository is partiotioned into the following files:

• The Pipeline folder contains the proposed algorithm along with all the required Python scripts for it to run. The scripts contain their own documentation.
  • The Results sub-folder contains the outputs of the proposed algorithm.
• The requirements.txt contains the required libraries to run the demo.
• The commands.sh contains the command line script used to run the testing suite as a whole: it first install the libraries, then executes the testing and finally aggregates the outputs into their own directory.

License

Distributed under the MIT License. See LICENSE.txt for more information.