ML-based prediction of elastic properties using reduced datasets of accurate calculations results

This repository contains data, described in the paper "Machine learning-based prediction of elastic properties using reduced datasets of accurate calculations results"

---

Files

• model_parametes.info - used parameters for model training
• custom_model.py - python library containing the model as a class object
• multimodel.py - script for using a pre-trained model
• test.txt - example input file for the multimodel.py
• data/db.feats.pt - initial train set, with pre-calculated features, described in the article
• data/Default_PAW_potentials_VASP.csv - file used to calculate features to use the model
• data/hull_feats.json - file used to calculate features to use the model
• data/model.dump - pretrained model

---

The main concept of this work is creation of two stacked estimators trained in a specific way. The first one is trained on large datatset of less accurate calculations made using EMTO-CPA and the second is trained on the much smaller dataset of more accurate PAW-SQS calculations.

---

Usage

• Ensure that all requirements reached (see requirements.txt
• python multimodel.py -i test.txt
• output will be saved as [input filename].csv