A deep neural network predicting melting temperature of alloy with arbitrary composition
Download the files and put them in a folder. In the folder, type:
tar zxvf liq.tgz
The code is dependent on pymatgen, PyTorch and scikit-learn, which should be installed first.
For Bayesian optimisation, dragonfly (https://github.com/dragonfly/dragonfly) should be installed first.
Adjustable parameters and their explanations are listed in the beginning of each script.
python generate_data.py
This command reads liquidus data from the files in the liq/ folder. A folder with default name "cv" containing all the necessary data for training will be generated.
python train.py
python ensemble_train.py
python bo.py
The configuration file should be present in the folder, and the default name is "config1.json", which is adjustable. The hyperparameters and associated loss function value in each evaluation will be printed.
This software was written by Pin-Wen Guan who was advised by Prof. Venkat Viswanathan.
Please cite the reference below if you use MeltNet in your work:
Pin-Wen Guan and Venkatasubramanian Viswanathan. “MeltNet: Predicting alloy melting temperature by machine learning”. In: arXiv preprint arXiv:2010.14048 (2020).
@article{guan2020meltnet,
title={MeltNet: Predicting alloy melting temperature by machine learning},
author={Guan, Pin-Wen and Viswanathan, Venkatasubramanian},
journal={arXiv preprint arXiv:2010.14048},
year={2020}
}