NNTuck

This github contains code and definitions for running NNTuck as described in A factor model for multilayer network interdependence. The NNTuck_tutorial.pynb steps through the methods discussed and performs the NNTuck on a multilayer social support network for a village surveyed in Banerjee et al.'s (2013) Diffusion of Microfinance.

The code depends on the following packages and the version number for which it is reproducible is noted. numpy (version 1.22.2), tensorly (version 0.5.1), sklearn (version 0.23.2), and matplotlib (version 3.3.2). When sweeping over parameters $K$ and $C$ in NNTuck it is most efficient to run the sweep in parallel, which depends on joblib (version 1.0.1) and on os to make sure the parallel runs don't use too much CPU.

Note that the implementation of the multiplicative updates for NNTuck was done by updating the current tensorly implementation for nonnegative Tucker decomposition (which minimizes Frobenius loss instead of KL).