- First install the prerequisite libraries:
- PyTorch, ASE, Pymatgen, TPOT, Scikit-learn, Matminer
- Clone this library locally as:
git clone https://github.com/hitarth64/therml
- Optional installs (refer to optionals for more information):
- ROOST from https://github.com/CompRhys/roost
- CrabNet from https://github.com/anthony-wang/CrabNet
- We use two sets of data in this study:
- Thermoelectric property database from MRL @ UCSB
- Closed-loop experimental dataset
therml
directory:- Contains files, modules and data required for training accurate ML model to perform error-correction learning and rank materials
- Please refer to the in-directory README file for more information
therml/saved_models
directory:- Contains the checkpoint of our model with highest cross-validation score
therml/prior_models
directory:- All the prior-models trained using Magpie, Roost and CrabNet (refer to manuscript for definition of prior-models)
- Please refer to the in-directory README file for more information
-
You can perform the inference using:
python inference.py
- Modify the inquiry dataloader within
inference.py
to rank new material candidates
-
You can perform error-correction learning using:
python hpo_dense.py
- Enables you to perform hyperparameter search for the error-correction model.
- It is setup by default, to train and cross-validate on all the data collected until the last round (which is what we did)
-
If you encounter any problem, feel free to start a discussion in the Issues
@article{https://doi.org/10.1002/adma.202302575,
author = {Choubisa, Hitarth and Haque, Md Azimul and Zhu, Tong and Zeng, Lewei and Vafaie, Maral and Baran, Derya and Sargent, Edward H},
title = {Closed-loop Error Correction Learning Accelerates Experimental Discovery of Thermoelectric Materials},
journal = {Advanced Materials},
volume = {n/a},
number = {n/a},
pages = {2302575},
doi = {https://doi.org/10.1002/adma.202302575},
url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/adma.202302575},
TherML is released under the MIT License.