Efficient Calculation of Polynomial Features on Sparse Matrices

Andrew Nystrom, John Hughes

Abstract:

We provide an algorithm for polynomial feature expansion that both operates on
and produces a compressed sparse row matrix without any densification. For a
vector of dimension D, density d, and degree k the algorithm has time complexity
O(d^k * D^k) where k is the polynomial-feature order; this is an improvement by a factor d^k
over the standard method.

TL;DR: An algorithm to perform polynomial expansions on CSR matrices that scales with matrix density polynomially.

Keywords: Supervised Learning

Conflicts: google.com, ed.ac.uk, cs.brown.edu, umn.edu