k-Pareto Optimality-Based Sorting with Maximization of Choice

This directory contains the code in Python for the paper "k-Pareto Optimality-Based Sorting with Maximization of Choice", preprint is available here. The notebooks can be run directly in Google Colab. Alternatively, you can download the repo and run it locally on your machine.

To generate images demonstrating different characteristics of sorting by k-Pareto optimality, check plots-for-paper.ipynb notebook. Alternatively, use the following scripts:

1. plotting_scripts/selections_40pct.py - generation of Figure 5: Selections of the best 40% according to different sorting criteria.
2. plotting_scripts/diversity_vs_area.py - generation of Figure 6: Fraction of the best elements and diversity.
3. plotting_scripts/rarefication.py - generation of Figure 8a from Annex with rarefication example.
4. plotting_scripts/trade_off.py - generation of Figure 8b from Annex with trade-off and cone-based relation example.
5. plotting_scripts/PO_whisker.py - generation of Figure 12 from Annex with a selection obtained for sorting by Pareto fronts and $k$-Pareto optimality.
6. wealthy_countries - data and analysis for Figure 16 from Annex (wealthy and non-wealthy states).

To test the proposed genetic optimization algorithms PO-count and PO-prob, use Knapsack-k_Pareto_optimality.ipynb.

Check the code in deap/tools/emo.py for the implementations of PO-count (sortPO()) and PO-prob (sortPOprob()).

Requirements:

• use the version of deap from the current repository,
• numpy,
• matplotlib,
• jupyter,
• pandas,
• tikzplotlib, (required only to generate plots in tikz format)
• scipy,
• sklearn,
• Java for the efficient calculation of hypervolume, see Hypervolume.java.