Multiplex Network Reconstruction using Entropy Maximisation Algorithms

Code accompanying master's thesis.

Author: Roman Aleksandrov

Title: "Development of a Link Reconstruction Method for Multiplex Networks", (University of Amsterdam, 2021).

A pdf link will be added later.

Note that there is another Python package NEMtropy which is probably a better solution for out-of-the-box single-layer network reconstruction.

Data

Datasets used:

1. FAO Multiplex. [Homepage] [Paper] [Data]. Unzip the data into FAO_dataset folder to run experiments.
2. Dutch Business Network. Private access.

Algorithms

The following algorithms are implemented:

1. MaxEnt
2. Iterative Proportional Fitting (IPF) + modifications
3. Fitness-induced Directed Binary Configuration Model (f-DMCM) + modifications
4. Multiplexity fitting

Installation

Tested with Python 3.6 and 3.8, package list attached in requirements.txt.

Evaluation of single-layer reconstruction

Code of the experiments is available for the FAO Network only. The scripts save resulting plots and tables into output folder (you might need to create it in advance).

Running for some scpecific layer by id to assess quality and generate reconstruction heatmap:

python -m experiments.fao_reconstruct -l 31

Running for n random layers with s different random seeds to get score tables quickly:

python -m experiments.fao_reconstruct -n 10 -s 3

Running for all layers:

python -m experiments.fao_reconstruct -a

Once evaluation for all layers is done, creating plots for F1-scores and constraints is possible with:

python -m experiments.visualise_scores

Evaluation of multi-layer reconstruction (multiplexity fitting)

Similarly, the experiments with two-layer reconstruction using IPF + multiplexity fitting can be done repeated with:

python -m experiments.fao_multireconstruct --all

Other possible options are: --single, -n N.

In addition, visualisation script is included. Need to set target file and desired plots in the script code.

python -m experiments.visualise_multiplexity

Description of contents

Starting point is the reconstruction folder. Algorithm implementations are placed there.

experiments folder contains:

• fao_reconstruct.py -- single-layer reconstruction for the FAO network
• fao_multireconstruct.py -- multi-layer experiments
• visualise_*.py -- various visualisations appearing in the paper

fao_data.py, sampling.py -- data loading functions for the FAO network.

multiplex folder contains helper functions for multi-layer experiments.

fao_analysis contains scripts for descriptive analysis of the network.