unnet is a library that can be used to study edge-uncertainty in networks for different network analysis tasks. See related preprint https://arxiv.org/abs/2010.11546.

To study edge uncertainty we start from an initial network which can be a real network or synthetic network. You can then apply different types of edge uncertainty, such as Jaccard noise or node-label based noise which systematically remove or add edges. Finally change can be quantified for subsequent network analysis tasks.

In unnet we provide a network generator for the barabasi+ homophily model in generators.py. Different samplers can be found in samplers.py. As network analysis task we provide wrappers for common network centrality measures in centralities.py.

Example usage

In the notebooks folder we have included several jupyter notebooks which can be executed to reproduce the experiments in a few minutes. In particular the notebook submission_plots_BA.ipynb can be run without the data. If you plan on using the real world networks, the public available ones can be downloaded by executing

python get_datasets.py

which stores them in the notebooks folder. You can then also run the submission_plots_real.ipynb notebook.

Installation instructions

This software uses the graph tool library https://graph-tool.skewed.de/ which is relatively difficult to install on non linux operation systems (see https://git.skewed.de/count0/graph-tool/-/wikis/installation-instructions). After installing graph tool you can install this library like any other python package, depending packages are also installed along with it.

Installing conda

wget https://repo.anaconda.com/archive/Anaconda3-2020.11-Linux-x86_64.sh
chmod +rwx Anaconda3-2020.11-Linux-x86_64.sh
bash Anaconda3-2020.11-Linux-x86_64.sh

Now follow the installation instructions for anaconda. Afterwards reopen the console to make sure anaconda is active.

Installing graph tool

conda create --name conda_unnet -c conda-forge python=3.8 graph-tool
conda activate conda_unnet

Installing unnet

git clone https://github.com/Feelx234/unnet.git
pip install -e unnet

The installation should take less than a minute. If you also want to run the demo notebooks please install jupyter as outlined below.

pip install jupyter
python -m ipykernel install --user --name=conda_unnet
python -m jupyter notebook

Now inside your jupyter make sure you are using the conda_unnet kernel and you are good to go.

System requirements

The unnet package should run on any modern standard computer.

Software requirements

OS Requirements

This package has been tested on macOS and Linux:

• macOS: Big Sur (11.1)
• Linux: Ubuntu 20.04.1 LTS

Python Dependencies

The code was tested with Python 3.8.5. unnet depends on the Python scientific stack.

numpy
pandas
matplotlib
tqdm

Parts of it also depend on the just-in-time compiler numba we used version 0.50.1.