implementation for Adaptive Structural Fingerprints for Graph Attention Networks
most of the code are copy from the https://github.com/AvigdorZ/ADaptive-Structural-Fingerprint
However the code in that link can not be run successfully, so i make another implement which all the data and code are included.
conda create -n ADSF python=3.9
pip install -r requirements.txt
python train.py
the output is in nohup.out
743.pklis the best model.
interdata has all the intermediate data, which are called in the utils_nhop_neighbours.py.
you can add a extra 'or True' after these branch to generate these files by yourself but it would cost a really long time(may be 3-6 hours)
if you have some great idea that can make the code has a better style, don't hesitate to take a PR.