phisnet_fork

Fork of original PhiSNet code used for the work in CasNet. For further instructions see: https://github.com/rhjvanworkum/CasNet

old instructions

Here we provide the code for training our PhiSNet model. Package requirements (lower versions might work, but were not tested):

• python >= 3.7
• numpy >= 1.20.2
• torch >= 1.8.1
• cudatoolkit >= 10.2
• ase >= 3.21.1
• apsw >= 3.34.0.r1
• tensorboardX >= 2.2

After installing the necessary packages, follow the steps below to train it on the PBE/def2-SVP dataset for water.

1. Download the tarball for the PBE/def2-SVP dataset for water from http://quantum-machine.org/data/schnorb_hamiltonian/schnorb_hamiltonian_water.tgz

2. Unpack the tarball and put the file "schnorb_hamiltonian_water.db" in this folder

3. Convert the dataset to our format by running

python3 convert_db.py You should now have a file called "h2o_pbe-def2svp_4999.db"

4. Train PhiSNet by running

python3 train.py @args.txt