README.md

In JSzitas/kernel: Fast Kernel Approximations

kernel

kernel is a package for various kernel approximation methods - Random Fourier Features, Orthogonal Random Features, Structured Orthogonal Random Features (currently). The main aim is to develop a fast plugin package that will calculate the kernel function for the user, with optional extensions, such as kernel ridge regression, being used currently as a feature for testing. More functionality for direct applications will be added later.

Written entirely in C++, with intended frontends in R (native), Python (future).

Installation

You can clone the current directory - the C++ scripts currently compile only with Rcpp::sourceCpp() - once the package is closer to finish this will be automated. (Currently some of the functions might fail to compile due to frequent changes - if they do please submit an Issue. )

The methods are intended to work with Bigmatrix objects for working with larger than memory objects. Over the longer term, a different approach might be used, as bigmatrix no longer seems supported.

To do list (near future):

• Port R implementation of Nystrom to C++
• Add Quasi-Monte Carlo Fourier Features
• Wrap in a single function with several dispatches

JSzitas/kernel documentation built on Jan. 20, 2020, 2:01 a.m.