https://github.com/KangchengHou/gntk

Implementation of Graph Neural Tangent Kernel (NeurIPS 2019)
https://github.com/KangchengHou/gntk

Last synced: 19 days ago
JSON representation

Implementation of Graph Neural Tangent Kernel (NeurIPS 2019)

• Host: GitHub
• URL: https://github.com/KangchengHou/gntk
• Owner: KangchengHou
• License: mit
• Created: 2019-09-08T14:44:23.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2020-01-28T00:56:42.000Z (almost 5 years ago)
• Last Synced: 2024-08-01T17:24:44.241Z (4 months ago)
• Language: Python
• Homepage:
• Size: 4.84 MB
• Stars: 100
• Watchers: 6
• Forks: 20
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-graph-classification - [Python Reference

README

        
# Graph Neural Tangent Kernel: Fusing Graph Neural Networks with Graph Kernels

This repository implements Graph Neural Tangent Kernel (infinitely wide multi-layer GNNs trained by gradient descent), described in the following paper:

Simon S. Du, Kangcheng Hou, Barnabás Póczos, Ruslan Salakhutdinov, Ruosong Wang, Keyulu Xu. Graph Neural Tangent Kernel: Fusing Graph Neural Networks with Graph Kernels. NeurIPS 2019. [[arXiv]](https://arxiv.org/abs/1905.13192) [[Paper]](https://papers.nips.cc/paper/8809-graph-neural-tangent-kernel-fusing-graph-neural-networks-with-graph-kernels)

## Test run

Unzip the dataset file

```

unzip dataset.zip

```

Here we demonstrate how to use GNTK to perform classification on IMDB-BINARY dataset. We set the number of BLOCK operations to be 2, the number of MLP layers to be 2 and c_u to be 1.

Compute the GNTK gram matrix

```

mkdir out

python gram.py --dataset IMDBBINARY --num_mlp_layers 2 --num_layers 2 --scale uniform --jk 1 --out_dir out

```

Classification with kernel regression 

```

python search.py --data_dir ./out --dataset IMDBBINARY

```

Therefore we get the hyper-parameter search results at `./out/grid_search.csv`.

## Experiment for all datasets

To run the experiment described in our paper, please run `bash run_gram.sh` and `bash run_search.sh`

in order.