/Material_Informatics

Basic components to perform material informatics: modeling (GPR, KRR, XGB, NN, RF, linear, and ensemble learning of them), backward prediction, multi target screening, etc.

Primary LanguageJupyter NotebookMIT LicenseMIT

Material_informatics

Includes:

  • Clustering - K-means, DBSCAN, GMM --- scripts to choose close materials for better models.
    clustering

  • Regression Models - Gaussian process, Kernel ridge, XGBoost, neural net (Keras Tensorflow), random forest, etc. --- scripts to use better kernel or hyperparameters.

regression_models

  • Random sample generation - Brute force, and Gaussian mixture model --- with specific restrictions. GMM_sample

  • Backward prediction and screening - Combine all process to find solution --- allows multiple targets and show solution reliability as circle sized. 20211003_composition_solution Verify if standard deviation truely reflects dot size (large means small deviation) 20211003_display_reliability_in_BPR_model

  • Automatic outlier detection - Powerful tool to improve performance of regression models. Thank you again Kaneko sensei Meiji University. ( https://datachemeng.com/outlier_samples_detectionc_python/ )