The GeoTMI is a model-agnostic method to solve the practical infeasibility of high-cost 3D geometry in many other chemistry fields. The aim of GeoTMI is maximazation of the mutual information between high-cost 3D geometries, correspoding quntum chemical properties, and easy-to-obtain geometries.
- torch==1.12.1
- ase==3.21.1
- torch-geometric==2.0.4
- cudatoolkit==11.3.1
- torch_cluster==1.6.0
source install.sh
install.sh:
conda create -n REP -y
source activate REP
conda install -c conda-forge mamba -y
mamba install xtensor-r -c conda-forge -y
mamba install pytorch=1.12.1 torchvision torchaudio cudatoolkit=11.3 -c pytorch -y
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.12.1+cu113.html
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.12.1+cu113.html
pip install torch-geometric==2.0.4
pip install ase==3.21.1
pip install networkx
pip install torch_cluster -f https://data.pyg.org/whl/torch-1.12.1+cu113.html
pip install sympy
pip install pandas
pip install rdkit-pypi
- Please put files and ocp directory to corresponding original equiformer directory. ex) cp OC20/equiformer/oc20/trainer/* [EQUIFOMRER_ORIGIN_PATH]/oc20/trainer/
- source GeoTMI.yml
- Read QM9M/data/QM9_REAME.md
- cd QM9M/data
- python gdb2mmff.py
- python preprocessing_dataset.py