Scalarize

This repository contains the code for our papers:

1. Multi-objective optimisation via the R2 utilities (https://arxiv.org/abs/2305.11774)
2. Random Pareto front surfaces (https://arxiv.org/abs/2405.01404)
3. Scalarisation-based risk concepts for robust multi-objective optimisation (https://arxiv.org/abs/2405.10221)

At its core, the scalarize code provides a useful collection of utilities for scalarization functions, scalarization parameters and scalarized objective functions that can be used alongside standard routines from NumPy, SciPy and PyTorch.

Dependencies

This code was initially implemented with the following dependencies:

• Python 3.9
• BoTorch 0.8.1
• PyTorch 1.13.1
• GPytorch 1.9.1
• Linear-operator 0.3.0
• SciPy 1.9.3
• NumPy 1.23.5

Organization

• The experiments folder contains the scripts and configurations which are used to execute the experiments.
• The notebook folder contains the notebooks used to obtain the plots used our papers.
• The plot_experiments folder contains the notebooks used to plot the results of the experiments.
• The scalarize folder contains the code needed to execute these strategies.
• The visualize folder contains the streamlit apps that can be used to visualize the one-dimensional or two-dimensional slices of a Pareto front surface.