SY-GNN

PyTorch implementation of Symmetrical Graph Neural Network for Quantum Chemistry with Dual Real and Momenta Space

Requirements:

ase>=3.10.1

numpy

python>=3.6

tensorboard>=1.15.0

tensorboardx>=2.0

torch>=1.5.0

Demo

Database generation

We provides QM_debug.db, which is a subset of QM-sym dataset in this repo. If you do not want to use the QM_debug.db to run the model. You need to build your database from XYZ files.

Modify dbpath and xyzpath in the last line in the file netpack/datasets/qm_sym.py. The dbpath is the save path of the generated database file, and xyzpath is the folder storage XYZ files.

Training

You can use following code to run with demo small database from QM-sym: (on Linux system)

CUDA_VISIBLE_DEVICES=0 python -W ignore main.py train sygnn --cuda --data_size 16 ./database/QM_debug.db ./logs/QM/debug

It should finish in few minutes if you have an CUDA device, the program should be expected to terminated without any error. The output log file in the log folder.

If you want to run with your own dataset, following the README of QM-sym dataset to build your XYZ files, and specify the folder to your .xyz file in the qm_sym.py to generate new database file.

Reference

Some codes are from SchNetPack.

The data is from QM-sym database.