Skip to content

A deep neural network predicting melting temperature of alloy with arbitrary composition

License

Notifications You must be signed in to change notification settings

PinwenGuan/MeltNet

Repository files navigation

MeltNet

A deep neural network predicting melting temperature of alloy with arbitrary composition

Installation

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.

General instructions

Adjustable parameters and their explanations are listed in the beginning of each script.

Generate data

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.

Train MelNet

python train.py

Train MelNet using the ensemble method

python ensemble_train.py

Bayesian Optimization of hyperparameters

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.

Author

This software was written by Pin-Wen Guan who was advised by Prof. Venkat Viswanathan.

Reference

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}
}

About

A deep neural network predicting melting temperature of alloy with arbitrary composition

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages