adasamp-pareto 1.1

Adaptive bayes-sampling for multi-criteria optimization

Adaptive optimization algorithm for black-box multi-objective optimization problems with binary constraints on the foundation of Bayes optimization. The algorithm aims to find the Pareto-optimal solution of

\begin{equation*} max [ y(x) ] s.t. f(x) = feasible \end{equation*}

in an iterative procedure. Here, \(y(x)\) denotes the multi-dimensional goals and \(f(x)\) the binary feasibility of the problem (in the sense that certain design variables \(x\) lead to invalid goals). All technical details can be found in the paper “Adaptive Sampling of Pareto Frontiers with Binary Constraints Using Regression and Classification” (https://arxiv.org/abs/2008.12005).

Installation

Install via pip or clone this repository. In order to use pip, type:

$ pip install adasamp-pareto

Usage

The class AdaptiveSampler is used to define and solve a problem instance. Simple example:

from adasamp import AdaptiveSampler

# Create instance
sampler = AdaptiveSampler(func,       # Problem definition: function returns (goals Y, feasibility f)
                          X_limits,   # Design variable limits to search solution in
                          Y_ref,      # Reference point, has to be dominated by any goal Y
                          iterations, # Number of solver iterations
                          Y_model,    # Regression model to predict goals Y
                          f_model)    # Classification model to predict feasibility f

# Return the sampling suggestions X, the corresponding goals Y, and the corresponding feasibilities f.
X, Y, f = sampler.sample()

Demo notebooks can be found in the examples/ directory.

Documentation

Complete documentation is available: https://adasamp-pareto.readthedocs.io/en/latest.

📖 Citation

If you find this code useful in your research, please consider citing:

@misc{heesebortzCITE2020,
              title={Adaptive Sampling of Pareto Frontiers with Binary Constraints Using Regression and Classification},
              author={Raoul Heese and Michael Bortz},
              year={2020},
              eprint={2008.12005},
              archivePrefix={arXiv},
              primaryClass={stat.ML}
     }