Awesome

GraphQNTK

This repository contains code implementation for the paper GraphQNTK: Quantum Neural Tangent Kernel for Graph Data, NeurIPS, 2022. Our implementation is based on the code(url) and we thank the authors for sharing.

Packages requirements

We use the following version of the packages to run the code, and a later version should also work properly.

networkx 2.2.1
numpy 1.19.5
scikit-learn 0.23.2
scipy 1.5.3

Datasets download

The datasets are available at this site. Download the compressed file and extract it into ./dataset under the root directory.

Train the model

To obtain the kernel matrix of the enhanced graph neural tangent kernel

python train.py --dataset MUTAG --num_mlp_layers 2 --num_layers 4 --scale uniform --out_dir out

The kernel matrix would be saved in thr file ./out/{dataset folder}/gram.npy. To make predictions based on the pre-defined kernel

python search.py --data_dir ./out/{dataset folder} --dataset MUTAG

The results would be found in ./out/{dataset folder}/grid_search.csv.

In order to simplify the estimation of the prediction results of different hyperparameters (such as different layers) under the same dataset, we provide a python script conclude.py for convenience. To output the results in one .csv file, run the code

python conclude.py --data_dir out --dataset MUTAG

The results would be saved in ./out/conclude-{dataset}.csv.