NFFTAddKer

This repository contains the implementation of the framework described in the paper "Fast Evaluation of Additive Kernels: Feature Arrangement, Fourier Methods, and Kernel Derivatives".

Usage

The main file FE_nfft_kernel_ridge.py consists of the following two classes:

• NFFTKernelRidgeFE performs a NFFT-accelerated KRR on additive kernels.
• GridSearch searches on candidate parameter values for the classifiers NFFTKernelRidge or sklearn KRR.

feature_engineering.py is the file in which all feature arrangement techniques are implemented.

Data sets

The benchmark data sets used in for the numerical results can be downloaded from the following websites: Protein, KEGGundir, Bike Sharing and Housing. The data files can also found in the data folder of this repository.

References

This repository uses the prescaledFastAdj package to perform NFFT-accelerated kernel evaluations for the Gaussian and the Matérn(1/2) kernels and their derivative kernels.

Citation

@article{wagner2024NFFTAddKer,
  title     = {Fast Evaluation of Additive Kernels: Feature Arrangement, Fourier Methods, and Kernel Derivatives},
  author    = {Theresa Wagner and Franziska Nestler and Martin Stoll},
  keywords  = {additive kernels, feature grouping, Fourier analysis, kernel derivatives, multiple kernel learning},
  url       = {https://arxiv.org/},
  year      = {2024}
}