Cartesian equivariant deep learning for molecular orbitals
Thank you for your interest in this work!
Source code for CEONet is avialable in src/deeporb, particularly src/deeporb/ceonet.py
Training scripts are found in scripts/training_scripts
Notebooks show example inference and figure generation, using the data in model_eval
- Clone the project
- Make a directory in the project called
deeporb/opt - Clone
cacedependency fromhttps://github.com/dking072/cace - Conda install the environment
- Activate the environment
Running in-memory:
- Convert
.h5file to.ptfile to store in memory.
python scripts/memory-experiments.py --file /eagle/DeepOrb/sto3g/subset/sto3g_occ_100000.h5 --convert
One time setup:
Make sure that the deeporb environment is installed in the project direectory
- Check the folder:
/eagle/DeepOrb/env/deeporb - If not installed
conda env create -f environment.yml —prefix=/eagle/DeepOrb/env/deeporb -n deeporb
- Make sure you add its location to your personal
.condarcfile
conda config --append envs_dirs /eagle/DeepOrb/env
Running training:
module use /soft/modulefiles
module load conda
conda activate deeporb
cd /eagle/DeepOrb/deeporb
python scripts/polaris/sto3g_occ_1000000.py
Interactive Job:
qsub -I -l select=1 -l filesystems=home:eagle -l walltime=1:00:00 -q debug -A DeepOrb
CACE (https://github.com/BingqingCheng/cace) ASE (https://wiki.fysik.dtu.dk/ase/install.html)