/srf

Random Fourier Features

Primary LanguageJupyter Notebook

Spherical Random Features - Review of (J. Pennington et al., 2015)

In this project Notebooks:

1- Random fourier features for Gaussian/Laplacian Kernels (Rahimi and Recht, 2007)

  • RFF-I: Implementation of a Python Class that generates random features for Gaussian/Laplacian kernels.
  • RFF-II: MSE evaluation of kernel matrices on USPS and Gisette datasets.
  • RFF-III: SVM accuracy / computation time statistics on USPS/Gisette using Gaussian kernel.

2 - Spherical Random Fourier for Polynomial kernels (J. Pennington et al., 2015)

Implementation as two classes, one for approximating the sampling PDF and another to sample Fourier Features.

  • SRF-I: Implementation of the first class ApproxKernel
  • SRF-II: Implementation of the second one SRFF.
  • SRF-III: SVM/MSE on USPS and Gisette
  • SRF-stats: Reproduction of all J. Pennington et al. results, tables and figures

All python classes are stored in .py files for more convenience.