November 1, 2019

204 words 1 min read



PyTorch Implementation of GraphTSNE, ICLR19

repo name leowyy/GraphTSNE
repo link
language Jupyter Notebook
size (curr.) 22555 kB
stars (curr.) 74
created 2019-04-15
license MIT License


Blog Post | Paper

GraphTSNE: A Visualization Technique for Graph-Structured Data International Conference on Learning Representations 2019 Workshop for Representation Learning on Graphs and Manifolds


The code demo_notebook.ipynb creates a visualization of the Cora citation network using GraphTSNE. The original Cora dataset and other citation networks can be found here:

The notebook takes roughly 3 minutes to run with GPU, or 8 minutes with CPU.


# Install Python libraries using conda
conda env create -f environment.yml
conda activate graph_tsne
python -m ipykernel install --user --name graph_tsne --display-name "graph_tsne"

# Run the notebook
jupyter notebook

When should I use this algorithm?

For visualizing graph-structured data such as social networks, functional brain networks and gene-regulatory networks. Concretely, graph-structured datasets contain two sources of information: graph connectivity between nodes and node features.


If you use GraphTSNE in your work, we welcome you to cite our ICLR'19 workshop paper:

  title={GraphTSNE: A Visualization Technique for Graph-Structured Data},
  author={Leow, Yao Yang and Laurent, Thomas and Bresson, Xavier},
  booktitle={ICLR Workshop on Representation Learning on Graphs and Manifolds},
comments powered by Disqus