This repository contains data, described in the paper "Machine learning-based prediction of elastic properties using reduced datasets of accurate calculations results"
model_parametes.info
- used parameters for model trainingcustom_model.py
- python library containing the model as a class objectmultimodel.py
- script for using a pre-trained modeltest.txt
- example input file for themultimodel.py
data/db.feats.pt
- initial train set, with pre-calculated features, described in the articledata/Default_PAW_potentials_VASP.csv
- file used to calculate features to use the modeldata/hull_feats.json
- file used to calculate features to use the modeldata/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.
- Ensure that all requirements reached (see
requirements.txt
python multimodel.py -i test.txt
- output will be saved as
[input filename].csv