NJmat

NJmat

(updated 2024/05/07)

Data-driven machine learning software for materials science and engineering

Particularly helpful for experimentalists.

NJmat: main NJmat module (word vectors word2vec/tsne language model, material featurization, machine learning, deep learning, genetic algorithm, SHAP etc.)

NJmatNLPSetup: Language model submodule to predict materials using MatBERT language model.

Language_models: MatBERT and word2vec (millions or 50,000 paper abstracts) models.

NJmatCDE: NLP (natural language processing) submodule to extract proper chemical formula/name from csv file.

NJmatCHG: Submodule using machine learning potential for materials property prediction (CHGnet)

NJmatVIS: Visualizer submodule

Please check datasets.rar for template .csv formats (train/test and prediction). The dataset templates (4 cases) are available in https://github.com/huangyiru123/NJmat_dataset.

Please cite: https://doi.org/10.1021/acs.jcim.4c00493

More Citations:

1. https://doi.org/10.1021/acs.jcim.4c00493
2. https://doi.org/10.1063/5.0064875
3. https://doi.org/10.1088/1361-648X/ac3e1e
4. https://doi.org/10.1021/acsaem.2c04066
5. https://doi.org/10.1021/acsami.2c00564
6. https://doi.org/10.1021/acsami.2c00568