This repository contains an implementation of the method introduced in the paper "A Preconditioned Interior Point Method for Support Vector Machines Using an ANOVA-Decomposition and NFFT-Based Matrix-Vector Products"
We propose employing a NFFT-accelerated matrix-vector product using an ANOVA decomposition for the feature space within a preconditioned interior point method for support vector machines. For more details, see the above-mentioned paper.
This package uses the FastAdjacency package by Dominik Bünger to perform NFFT-based fast summation to speed up kernel-vector multiplications for the ANOVA kernel.
- This software has been tested with Python 3.8.
- This software depends on Bünger's FastAdjacency Package. We refer to https://github.com/dominikbuenger/FastAdjacency#readme for installation instructions.
The main file class_NFFTSVMipm.py consists of the following two classes:
NFFTSVMipmperforms a preconditioned interior point method for Support Vector Machines.RandomSearchsearches on random candidate parameter values for one of the classifiersNFFTSVMipmorLIBSVM.
It can be run via the showcase file run_NFFTSVMipm.py.
The benchmark datasets used in the numerical results section can be downloaded from the following websites: HIGGS, SUSY, cod-rna. The cod-rna data files can be found in the data folder of this repository. The HIGGS and SUSY data files exceed the standard size limits of GitHub and should be saved locally in the data folder of this repository.
We refer to our previous repository NFFT4ANOVA, where this fast NFFT-based matrix-vector product approach is applied to kernel ridge regression.