Code and Data Repository for Bi-fidelity Surrogate Modelling: Showcasing the Need for New Test Instances

Abstract

The code implements Kriging and Co-Kriging as surrogate models. That is, the implementation generates a sample of a bi-fidelity function (a high fidelity sample and a low fidelity sample of specified size), and trains the chosen surrogate model, without the added functionality of choosing further samples. Both of these models are described in the appendix given in the appendix folder. Multiple literature test instances are implemented here, as well as a novel instance creation procedure which adds a disturbance to a high-fidelity function to create a low-fidelity function. In this implementation functions from the COCO test suite16 are used as the high fidelity functions. The performance of both Kriging and Co-Kriging is assessed by calculating the Relative Root Mean Squared Error (RRMSE) between the trained model and the high fidelity source.