This repository was created as part of my Master thesis: Transfer learning of interatomic potentials: from molecules to crystals.
The aim was to use ANI2x model and apply it to crystals as a part of the Seventh CCDC Blind Test of Crystal Structure Prediction Methods. Structures can be found here: https://www.ccdc.cam.ac.uk/Community/initiatives/cspblindtests/7-csp-blind-test-targets/ .
Cuda 11.1 compatible graphics card
docker run --gpus all -ti --ipc=host -v /PATH/TO/FOLDER:/app nvidia/cuda:11.2.0-devel-ubuntu20.04
apt update
apt install python3-pip
pip3 install torch==1.9.0+cu111 -f https://download.pytorch.org/whl/torch_stable.html tqdm ase torchani seaborn
Datasets are assumed to be in the num.traj format in folder data, where num represents the structure number.
python3 generate_dataset.py -n num.traj
Takes the num.traj file and creates the /data/train_data_num.pkl file which can be fed directlly to model.
python3 train.py -n num
model is saved at model/compiled_model_num.pt directory.
python3 train.py --help
python3 test.py -n num
python3 model2cpu.py -n num
python3 data_visualisation.py -n num