Machine learning code and feature generation for crystal defect identification
The code generates all training data and trains a dataset based on parameters entered into generate_defect_structs.py It is based on having several lammps dump templates constructed with the defects of interest, and the defective atoms are then extracted using either CSP or CNA for training. After the data is generated it can be either saved or loaded in generate_defect_structs, where the actual training occurs. After the classification algorithms are trained they are saved in an extrernal folder for later use.
Each new identification is performed with the identify.py file. The command line parameters given to the file tell it what to identify with which ML algorithm.