Combining graph neural networks and classical kernels for graph generation

Pytorch implement of the article Combining graph neural networks and classical kernels for graph generation.

This library refers to the source code https://github.com/yongqyu/MolGAN-pytorch

Dependencies

1. python>=3.5
2. pytroch>=1.10.0 https://pytorch.org
3. rdkit https://www.rdkit.org/
4. numpy
5. sklearn

Setup

Before running:

1. Enter the data folder and run download_dataset.sh or extract the files at gdb9.tar.gz in the data folder
2. From the data folder run:

python sparse_molecular_dataset.py

Run the code

python main.py --name runname

Will result with the run's logs at /results/runname/logs.txt
Samples at /results/runname/fake_samples
Models at /results/runname/models

Hyper parameters

For information about the hyper parameters run:

python main.py --help

You can also view this instructions.
To recreate the experiments described in this article view this instructions.