drkupi / gomors_pysot

GOMORS - Efficient surrogate global optimization method for Multi-Objective global problems

GOMORS_pySOT

GOMORS is a surrogate-assisted Multi-Objective Optimization (MO) strategy, designed for computational expensive MO problems, e.g., expensive environmental simulation optimization problems, hyperparameter tuning of Deep Neural Networks etc. GOMORS is implemented in the pySOT lbirary and framework, and uses Radial Basis Functions (RBFs), as surrogates. Moreover, GOMORS uses a Multi Objective Evolutionary Strategy (MOEA) to optimize RBF surrogates in each iteration. Any MOEA methods can be connected with GOMORS for optimizing surrogates. We currently use the Platypus library to link GOMORS with epsion-MOEA. GOMORS also supports modest parallelization on up to 4 cores, and hence, is suitable for deskptop and laptop machines.

Installation Instructions

The prerequisites for using the GOMORS code in this repository is installation of a python 2 environment, the pySOT library, the platypus library and matplotlib. We recommend using a virtual environment within Anaconda. Instructions for installation of pre-requisites is as follows:

conda create --name mo-surrogate python=2.7
conda activate mo-surrogate
pip install pysot==0.1.36
pip install matplotlib
pip install platypus-opt

Running GOMORS

An example of how to run GOMORS is provided in the file simple_experiment.py. The setup for running the algorithm is synonymous to how optimization experiments are setup in pysot. To link GOMORS to a user-defined MO optimization problem, kindly look at how problems are defined in the test_problems.py python file. For further information please write to me at [email protected].

Toy SWAT Model

GOMORS can also be used for watershed model calibration. A toy SWAT model is provided in "tbrook" subfolder. Moreover, an example of running an MO SWAT watershed model calibration using GOMORS is provided in water_experiments.py.

Citing GOMORS

If you use GOMORS, please cite our paper, Akhtar, T., Shoemaker, C.A. Multi objective optimization of computationally expensive multi-modal functions with RBF surrogates and multi-rule selection. J Glob Optim 64, 17–32 (2016). https://doi.org/10.1007/s10898-015-0270-y.