Synthetic Graph Generation for DGL-PyTorch

NVIDIA Deep Learning Examples

Resource

NVIDIA Deep Learning Examples

Synthetic Graph Generation for DGL-PyTorch

The Synthetic Graph Generation tool enables users to generate arbitrary graphs based on provided real data.

Our results were obtained by running the demo notebooks directory in the PyTorch NGC container on NVIDIA DGX1 V100 with 8x V100 32GB GPUs. All the notebooks are presented in the table below.

	scope	notebook	description
1.	basic_examples	er_demo.ipynb	generating different types of random graphs using Erdős–Rényi model
2.	basic_examples	ieee_demo.ipynb	generating a bipartite graph structure based on provided edge list
3.	basic_examples	cora_demo.ipynb	generating a non-bipartite graph structure based on provided edge list
4.	advanced_examples	e2e_cora_demo.ipynb	a complete process of reconstructing and analyzing a non-bipartite graph dataset with node features
5.	advanced_examples	e2e_ieee_demo.ipynb	a complete process of reconstructing and analyzing a bipartite graph dataset with edge features
6.	advanced_examples	frechet_lastfm_demo.ipynb	SynGen non-bipartite graph structure generators scaling analysis
7.	advanced_examples	frechet_tabformer_demo.ipynb	SynGen bipartite graph structure generators scaling analysis
8.	advanced_examples	edge_classification_pretraining.ipynb	using synthetic data from SynGen for GNN pretraining
9.	performance	struct_generator.ipynb	comparison of SynGen graph structure generators
10.	performance	tabular_generator.ipynb	comparison of SynGen tabular data generators

Scope refers to the directories in which the notebooks are stored and the functionalities particular notebooks cover . There are

Basic - basic_examples - notebooks with the examples of basics functionalities
Advanced - advanced_examples - notebooks with the examples of advanced functionalities
Performance - performance - notebooks with the performance experiments

To achieve the same results, follow the steps in the Quick Start Guide.

Results

1. Quality of the content of generated dataset vs. original dataset:

The quality of the content comparison was conducted on the IEEE dataset (refer to List of datasets for more details) with corresponding notebook e2e_ieee_demo.ipynb We compared three modalities, that is, quality of generated graph structure, quality of generated tabular data and quality of aligning tabular data to the graph structure.

Graph structure quality
- Comparison of degree distribution for an original graph, properly generated and random (Erdős–Rényi)
- Comparison of basic graph statistics for an original graph, properly generated and random (Erdős–Rényi) ![graph_structure statistics](https://raw.githubusercontent.com/NVIDIA/DeepLearningExamples/master/Tools/DGLPyTorch/SyntheticGraphGeneration//img/graph_structure statistics.png)
Tabular data quality
- Comparison of two first components of a PCA of real and generated data
- Comparison of basic statistics between real and generated data
  
  Generator kl divergence correlation correlation
  GAN 0.912 0.018
  Gaussian 0.065 -0.030
  Random 0.617 0.026
Structure to tabular alignment quality
- Degree centrality for feature distribution

Generator	kl divergence	correlation correlation
GAN	0.912	0.018
Gaussian	0.065	-0.030
Random	0.617	0.026

2. Performance (speed) of the synthetic dataset generation:

Performance of graph structure generation (edges/s)
Performance of categorical tabular data generation (samples/s)

Dataset KDE KDE_SK Uniform Gaussian CTGAN
ieee 1420600 2640973 1768136 1509729 33202

Dataset	KDE	KDE_SK	Uniform	Gaussian	CTGAN
ieee	1420600	2640973	1768136	1509729	33202

3. Synthetic dataset use-case specific quality factors:

Performance (batches/s) comparison between original vs. synthetic datasets

Dataset Model Synthetic Original
ieee gat 0.07173 0.07249