Dataset Generation and Model Testing

This repository contains the code for generating three kinds of datasets (ULBD, CDDD, GCDD) and for testing the model trained on ULBD, alongside a comparison with CDDD.

1. Dataset Generation

1.1 Generate CDDD

Steps to generate CDDD:

1. Get CIF from Materials Project

   • Run CDDD_1_get_Cif.py to obtain the CIF files from the Materials Project.
   • Note: You need to obtain your own API key from the Materials Project website.

2. Generate Diffraction Patterns

   • Run CDDD_2_get_Dataset.py to generate the diffraction patterns from the CIF files obtained in Step 1.

---

1.2 Generate GCDD

Steps to generate GCDD:

1. Generate CIF using Pyxtal

   • Run GCDD_1_generate_Cif.py to create the CIF files.

2. Generate Diffraction Patterns

   • Run GCDD_2_get_Dataset.py to generate the diffraction patterns from the CIF files obtained in Step 1.

3. Clean the Dataset

   • Run GCDD_3_cleaning_Dataset.py to remove any NaN and infinite values generated in Step 2.

---

1.3 Generate ULBD

Steps to generate ULBD:

• Run ULBD_get_Dataset_*.py to generate the ULBD for the specific crystal system you are interested in.

---

2. Model Testing

2.1 Validate Predictions Against Extinction Law

Steps to validate predictions:

• Run test_cubic/tetragonal/trigonal_hexagonal.py to obtain the top-1 to top-5 accuracies of the model.

---

2.2 Test Generalization Ability of Models

Steps to test generalization ability:

• Run test_cubic/tetragonal/trigonal_hexagonal_label_by_extinction.py to evaluate predicting accuracies.
• Change the test set and model used in the scripts to observe the variations in accuracy.

---

Note: Make sure to follow the instructions carefully and adjust any necessary parameters in the scripts as per your requirements.