desdeo 0.1.3

Open source library for for interactive multiobjective optimization

DESDEO README

DESDEO is a free and open source Python-based framework for developing and experimenting with interactive multiobjective optimization.

Documentation is available.

Background and publications available on the University of Jyväskylä Research Group in Industrial Optimization web pages.

Installation

From PyPI using pip

You can install desdeo from PyPI by running the following command in a terminal:

pip install desdeo

Background

There exist many methods to solve multiobjective optimization problems. Methods which introduce some preference information into the solution process are commonly known as multiple criteria decision making methods. When using so called interactive methods, the decision maker (DM) takes an active part in an iterative solution process by expressing preference information at several iterations. According to the given preferences, the solution process is updated at each iteration and one or several new solutions are generated. This iterative process continues until the DM is sufficiently satisfied with one of the solutions found.

Many interactive methods have been proposed and they differ from each other e.g. in the way preferences are expressed and how the preferences are utilized when new solutions. The aim of the DESDEO is to implement aspects common for different interactive methods, as well as provide framework for developing and implementing new methods.

Architecture

Overview of the current DESDEO architecture is shown in diagram

Interactive Methods

NAUTILUS Method

Most interactive methods developed for solving multiobjective optimization problems sequentially generate Pareto optimal solutions and the decision maker must always trade-off to get a new solution. Instead, the family of interactive trade-off-free methods called NAUTILUS starts from the worst possible objective values and improves every objective function step by step according to the preferences of the decision maker. Recently, the NAUTILUS family has been presented as a general NAUTILUS framework consisting of several modules. To extend the applicability of interactive methods, it is recommended that a reliable software implementation, which can be easily connected to different simulators or modelling tools, is publicly available. In this paper, we bridge the gap between presenting an algorithm and implementing it and propose a general software framework for the NAUTILUS family which facilitates the implementation of all the NAUTILUS methods, and even other interactive methods. This software framework has been designed following an object-oriented architecture and consists of several software blocks designed to cover independently the different requirements of the NAUTILUS framework. To enhance wide applicability, the implementation is available as open source code.

Examples

The functioning and flexibility of the DESDEO framework is demonstrated with two numerical example problems.

Development

Set-up

You should install the git pre-commit hook so that code formatting is kept consistent automatically. This is configured using the pre-commit utility. See the installation instructions.

Tests

Tests use pytest. After installing pytest you can run:

pytest tests

Release process

1. Make a release commit in which the version is incremented in setup.py and an entry added to HISTORY.md

2. Make a git tag of this commit with git tag v$VERSION

3. Push -- including the tags with git push --tags

4. Upload to PyPI with python setup.py sdist bdist_wheel and twine upload dist/*

Documentation

The full documentation is located at https://desdeo.readthedocs.io/en/latest/

Information about the academic project, including publications is available at http://desdeo.it.jyu.fi

History

0.1.3 (2018-06-18)

• Fix PyPI package

0.1.2 (2018-06-13)

• Improvements to automatically generated documentation
• Move river pollution example into desdeo.problem.toy module
• Allow solutions as well as objective functions to be obtained by adding ResultSet class
• Improvements to NIMBUS
  • Add missing NIMBUS scalarising functions
  • Add method to generate in-between solutions for NIMBUS
• Add RangeEstimator module for finding the nadir/ideal with a payoff table

0.1.1 (2018-05-21)

• Convert package description to use Markdown

0.1.0 (2018-04-25)

• First release