I have an input stream that consists of floats. I have a std::vector< std::vector<float> >, because the data is really several columns of numbers, each of which I want to keep seperate.

This is the implementation of the paper Lossless Compression of Vector IDs for Approximate Nearest Neighbor Search by Daniel Severo, Giuseppe Ottaviano, Matthew Muckley, Karen Ullrich, and Matthijs ...