Code accompanying the paper "Cepstral k-means"
The Python code in this repository recreates the numerical examples in the manuscript "Cepstral k-means", a preprint of which can be found on arXiv, by Oliver Lauwers and Bart De Moor, which develops a novel time series clustering algorithm, providing a rigorous framework for clustering time series according to their underlying dynamics. It combines the well-known k-means algorithm with the weighted cepstral distance. This manuscript has been sent in to be considered for publication to the Journal of Machine Learning Research.
Two main applications are available:
- An example using Cylinder-Bell-Funnel data, in a .py-file. (CylinderBellFunnelExample.py)
- A Jupyter notebook on unsupervised wind turbine generator anomaly detection. (Wind Turbine.ipynb)
Bear in mind that this code is not meant as a fully working software package, but serves merely as an illustration accompanying the manuscript mentioned earlier.
When using this code or discussing results of the cepstral k-means algorithm, please refer to this paper.
To refer to the data presented here: it comes from the Hack The Wind competition by EDF: https://opendata.edp.com/pages/hackthewind . It was published under a CC BY-SA 4.0 license ( https://creativecommons.org/licenses/by-sa/4.0/ ), allowing for sharing and adapting the data.