This repository contains the implementation of the Event-Triggered Kernel Adjustment for Gaussian Process Modelling (ETKA) algorithm. The corresponding publication will be referenced here after acceptance.
The code builds on pytorch, gpytorch and botorch to offer a general-purpose framework for Gaussian Processes. It also includes the Compositional Kernel Search and the Adjusting Kernel Search. The provided data is separated into synthetic data generated with Nike's time series generator and real-life data published by Lloyd et al..
The Research folder contains all data, data statistics and results used for the publication. It also includes a jupyter notebook with the code to reproduce all plots we show.
The implementation was done by Jan David Hüwel. The data generation was done by Florian Haselbeck. The overall research was conducted by them both together with Dominik G. Grimm and Christian Beecks.