This repository contains code for producing graph signals from METIS recursive bipartitions (recursive.py). The bipartitions establish dyadic scales of signals, with $\pm 1$ indicators splitting the "new" subgraphs at each layer. To build a full set of signals spanning a range of frequencies, the $i^{th}$ signal is formed as the product of the dyadic-scale signals indicated by the ones in the binary expansion of $i$. Then the signals are grouped by their highest dyadic scale (a coarse measure of signal frequency) and sorted within each group on their Rayleigh quotients with the graph Laplacian.

Loading a general collection of graphs can be done using make_hgr_general.py, with partitioning performed by recursive.py. Both of these files have command-line interfaces which are laid out at their respective bottoms. Importing benchmark graphs from an external source, such as OGBG-MolHIV or OGBG-PPA or ZINC12k, can be done using files in the specialty_loaders directory.