Practical overview
When it fits
Use Simgnn for research or experimentation when you have graph-pair data and want a documented PyTorch implementation of learned graph similarity.
Who it is for
- Researchers experimenting with neural graph similarity or graph edit distance.
- Developers preparing labeled graph-pair datasets for model training and testing.
Documented uses
- Represent graph pairs as JSON training or test examples.
- Configure a graph-similarity model through the training script's command-line options.
- Explore learned similarity scores for graph-comparison research.
Requirements & limits
The public GitHub repository is identified as open source under GPL-3.0.
- The documented input format requires paired graphs, node-label arrays and an integer GED value.
- The README specifies Python 3.5.2 and older package versions, including PyTorch 1.1.0 and PyTorch Geometric 1.3.2.
- The model exposes configuration for graph convolution layers, batch size, epochs, learning rate, dropout and optional histogram features.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare graph-pair JSON data
Structure training or testing examples in the format accepted by the repository.
Before you start
- Graph pairs for training or testing.
- Each graph pair represented as a JSON file.
Steps
Represent each graph's connectivity as an edge list under graph_1 and graph_2.
Add labels_1 and labels_2 arrays, with positions corresponding to node identifiers.
Include the raw integer graph edit distance under ged.
Index every node identifier and node label from zero, and store dictionary keys as strings for JSON serialization.
Expected outcome
A graph-pair JSON file in the input structure described by the README.
Guide limitations
- The guide covers data formatting only; it does not validate a dataset or run training.
Configure the documented training inputs
Select training and testing graph folders and review model settings exposed by the training script.
Before you start
- Python 3.5.2 and the package versions listed in the README, including PyTorch 1.1.0 and PyTorch Geometric 1.3.2.
- Training and testing folders containing graph-pair JSON files in the documented format.
Steps
Use the repository's src/main.py training script.
Set the training and testing graph folders with --training-graphs and --testing-graphs, or use the documented defaults dataset/train/ and dataset/test/.
Review the exposed model options, such as --filters-1, --filters-2, --filters-3, --batch-size, --epochs, --learning-rate, --dropout and --histogram.
Expected outcome
A documented configuration of the training inputs and available model parameters.
Guide limitations
- The README lists development package versions; it does not establish compatibility with other versions.
- This guide covers configuration options, not a verified training run.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 6 November 2026
- Original website title
- benedekrozemberczki/SimGNN · Official homepage title · retained in its source language
Related options
Reference TensorFlow SimGNN implementation — The README links to a separate TensorFlow implementation of SimGNN.