はてなブックマーク

by Erich Owens Optimizing the memory footprint of a classifier used here at Newsle set us down a rabbit hole of rewriting a basic Scipy function with Cython, something that only became a problem when our high-dimensional text spaces grew to a cartoonish size, thanks to the hashing trick. Here I motivate the use of the hashing trick, how we use sparse matrix-vector multiplication for text classific