Spectroscopy Masterclass Tutorial: From XAS to Machine Learning

This repository contains materials from the Spectroscopy Masterclass tutorial delivered at the CECAM workshop, held at the Rutherford Appleton Laboratory, UK (home of the Diamond Light Source synchrotron).

The tutorial provides a hands-on, end-to-end workflow for going from raw X-ray Absorption Spectroscopy (XAS) data to trustworthy machine learning models.

🗂 Contents

• Notebooks
  • Database
  • Data quality checks and preprocessing
  • Comparison with different XAS simulation codes
  • Generating simulated spectra with FDMNES
  • Feature engineering for spectroscopy data
  • ML pipelines with CDF+XGBoost, and PCA+MLP/1D-CNN
  • Model validation and transfer to experimental spectra
  • Hyperparameter tuning

📦 Dependencies

Required

• numpy
• matplotlib
• scikit-learn
• scipy
• xgboost (on macOS also install libomp; with conda: conda install -c conda-forge libomp xgboost)
• torch (PyTorch)
• pymatgen

Optional (for code in the markdown cells)

• optuna or scikit-optimize (Bayesian HPO)
• shap (feature attributions for tree models)