Graph Neural Networks for Graph Drawing

This repository contains the implementation for the paper Graph Neural Networks for Graph Drawing, accepted for publication in the IEEE Transactions on Neural Networks and Learning Systems. Technical Report

Authors: Matteo Tiezzi, Gabriele Ciravegna and Marco Gori.

Make sure to have Python dependencies by running:

pip install -r requirements.txt

We tested the code with PyTorch 1.10 and Deep Graph Library (DGL). Follow the instructions on the official websites for further details.

REPOSITORY DESCRIPTION

The folder structure is the following:

data :                                   folder with dataset and utilities to generate datasets
viz_utils:                               folder containing utilities/loss functions                
crossing_dataset_creator.py :            utility to create the dataset for training the Neural Aesthete on edge-crossing
crossing_learning_mlp.py :               script to train the Neural Aesthete
crossing_test_algorithm.py :             utilities for the Neural Aesthete
gd_stress.py :                           utilities to Draw Graphs using Graph Drawing force-directed packages ("neato", "pivotmds", "forceatlas2") 
graph_draw_main.py :                     script to draw graphs by iteratively optimizing a loss function with SGD (standard GD approaches). 
graph_neural_drawers.py :                script to train GNNs for Graph Drawing (datasets from the paper)
inference_gnn_stress_huge.py :           script to draw huge graphs with a pretrained Graph Neural Drawer
random_graph_factory.py :                script to generate the Sparse dataset from the paper
split_rome_dataset.py :                  script to create train/val/test split (Rome dataset)

DATASETS

The Rome dataset can be found at this link and unpacked into the data folder.

The Sparse dataset can be generated with the random_graph_factory.py script.

To obtain train/val/test splits, please exploit the split_rome_dataset.py script (similar code can be used to split the other datasets).

HOW TO TRAIN THE NEURAL AESTHETE

Launch the crossing_dataset_creator.py script to create the synthetic dataset to train the Aesthete. Then, launch the crossing_learning_mlp script to train a Neural Aesthete. The model will be saved in the saved_models folder.

Standard Graph Drawing with Neural Aesthete

Use the graph_draw_main.py script to draw graphs (please refer to the Arguments and the paper for further details) using standard SGD and the Neural Aesthete.

Graph Neural Networks for Graph Drawing

Use the graph_neural_drawers.py script to draw graphs (please refer to the Arguments and the paper for further details ) using GNNs and the Neural Aesthete.

Draw graphs with Graph Drawing packages

Please install the followig dependendencies before running the gd_packages.py script:

pip install networkx graphviz  pydot
pip install networkit 
# for windows 
winget install graphviz

# install ForceAtlas2  
git clone https://github.com/cvanelteren/forceatlas2.git  
python setup.py install