pmf is a python library for performing inference using hierarchical Poisson Matrix Factorization, see Graph link prediction in computer networks using Poisson matrix factorisation.

pmf requires python 3.7 or higher.

To install pmf from the source code

git clone https://github.com/mjmt05/pmf.git
cd pmf
pipenv install .

Example python script for training a model is provided in the examples folder

./train.py -h

An example edge list file is also provided, to run the script with default arguments

./train.py -f train.txt

simulation_test has a script for simulating from the model, running this will simulate a training and test data set from the model. It performs inference on the training data and assesses predictive performance on the test data set using the area under the ROC curve.

Use the python script in regression_test to validate any changes to the code. Add to the test when implementing new features.

Please use the following bibtex for citing pmf in your research:

@article{Sanna:2021,
author = {Sanna, Passino F and Turcotte, MJM and Heard, NA},
journal = {Annals of Applied Statistics},
title = {Graph link prediction in computer networks using Poisson matrix  factorisation},
url = {https://arxiv.org/abs/2001.09456},
year = {2021}
}

This code is released under the MIT license.