INFOMNWSC: Community Measure Analysis

Reproducing Results

1. This codebase was developed on Python 3.9.1. For best results, use that version of Python or later.
2. To install the necessary packages, run pip install -r requirements.txt.
3. To reproduce the main results of the paper, run python assess.py. In addition to logging many intermediate results to stdout, this will save results of the experiment to two files: data/benchmark_results.csv (which contains aggregated statistics such as NMI mean and variance for each measure) and data/benchmark_results_full.csv (which contains the results broken down by each synthetic graph seed).

Navigating the Codebase

The core functionality, should you wish to inspect it, is spread across several files:

• The code that implements our LFR benchmark-inspired synthetic graphs is in graph_generation_fs.py.
• The code that runs the optimization algorithm on each combination of synthetic graph and measure is in assess.py.
• Our implementation of the Louvain algorithm is in algorithm/louvain.py.
• Our implementations of the community measures are found in algorithm/edge_ratio.py, algorithm/intensity_ratio.py, algorithm/modularity.py, and algorithm/modularity_density.py.

Other files contain functionality of various utility, but are not necessary to reproduce the results of the paper.

Godspeed!