DeepSorption

This repository is the official implementation of DeepSorption.

Requirements

python         3.7
matplotlib     3.5.1
numpy          1.21.2
pandas         1.3.4
scikit-learn   0.22.1
torch          1.7.1
tqdm           4.42.1
mendeleev      0.9.0

Dataset

We provide raw datasets and pre-processed datasets of material science under https://doi.org/10.5281/zenodo.7699719.

Demo

If you want to use our trained model directly for adsorption prediction, please excute cd train_and_predict and run: python CoREMOF/pred.py.

The preprocessed training, validation and test set for Coremof dataset are stored in save/coremof/COREMOF_train.npy, save/coremof/COREMOF_dev.npy, and save/coremof/COREMOF_test.npy, respectively. This may take about 2 minutes to complete and report the performance.

Training example

To train DeepSorption on CoREMOF dataset, please excute cd train_and_predict and run:

python CoREMOF/main.py

Prediction example

To test the model performance on CoREMOF dataset, please excute cd train_and_predict and run:

python CoREMOF/pred.py

EXPMOF example

To train and test DeepSorption on EXPMOF dataset, please excute cd train_and_predict and run:

python EXPMOF/leave_one_out.py --expmof 'c2h2' or python EXPMOF/leave_one_out.py --expmof 'co2'

(Modify --expmof to specify different datasets.)