MINOAN

Mixed INteger Optimization using ApproximatioNs (Beta Version 0.0.1)

About

MINOAN is an open-source Python library used for machine learning-based (or surrogate-based) optimization. The alglorithm supports constrained NLP and MINLP (with binary variables) problems:

min f(x,y) 
s.t. g1(x,y)>=0, g2(x,y)<= 0, g3(x,y)=0
x_l <= x <= x_u, y = {0,1} 

It currently supports the following machine learning models:

• Artificial Neural Network (tanh and relu activation function)
• Gaussian Process
• Support Vector Regression

These models are constructed using scikit-learn and optimized using Pyomo via GAMS or NEOS interface. MINOAN has additional capabilities such as:

• Parallel processing for multiple promising binary solutions
• Gray-box problems with known/explicit constraints

If you have any questions or concerns, please send an email to sophiekim0205@gmail.com or fani.boukouvala@chbe.gatech.edu

Installation

If using Anaconda, first run: conda install git pip

The code can be directly installed from github using the following command: pip install git+git://github.com/DDPSE/minoan

Examples

Example codes are found in the directory "test".

• Example 1: constrained, black-box MINLP problem
• Example 2: constrained, gray-box MINLP problem
• Example 3: constrained, black-box NLP problem

References

• Kim SH, Boukouvala F. Machine learning-based surrogate modeling for data-driven optimization: a comparison of subset selection for regression techniques. Optimization Letters. 2019.
• Kim SH, Boukouvala F. Surrogate-Based Optimization for Mixed-Integer Nonlinear Problems. Computers & Chemical ENgineering. 2020.