materialyzeai/maml

Python for Materials Machine Learning, Materials Descriptors, Machine Learning Force Fields, Deep Learning, etc.

What it solves

Materials science researchers often struggle to bridge the gap between raw atomic structures and machine learning models. This project simplifies that process by providing a unified interface to convert complex material data into formats that standard machine learning tools can understand and use for predictions.

How it works

Instead of reinventing machine learning algorithms, this tool acts as a bridge. It leverages existing, powerful libraries like scikit-learn, TensorFlow, and MatGL to perform the heavy lifting. It provides specialized modules to turn atomic compositions and structures into numerical features, select diverse training data, and run advanced applications like interatomic potential fitting or Bayesian optimization for structure relaxation.

Who it’s for

This is designed for materials scientists and researchers who want to apply machine learning to their data without building custom infrastructure from scratch. It is ideal for those already working with tools like pymatgen or matminer who need a streamlined way to train models on material properties.

Highlights

  • Provides automated conversion of material structures into machine-readable features like Coulomb matrices and graph-network embeddings.
  • Includes specialized tools for fitting machine-learning interatomic potentials (such as SNAP, GAP, and MTP).
  • Features DIRECT sampling to help researchers select diverse, high-quality training sets from large datasets.
  • Offers pre-built applications for tasks like Bayesian optimization for structure relaxation and calculating grain boundary energies.
  • Integrates seamlessly with popular Python machine learning libraries while supporting advanced workflows like symbolic regression.

Related

  • Project
    materialyzeai/matgl

    MatGL is a graph deep learning library for materials science that enables the development and use of high-performance surrogate models to predict material properties and simulate atomic interactions.

  • Project
    Nixtla/mlforecast

    A scalable framework for time series forecasting that allows any scikit-learn compatible machine learning model to be used for forecasting millions of series.

  • Project
    feature-engine/feature_engine

    A Python library providing Scikit-learn compatible transformers for comprehensive feature engineering and selection to improve machine learning model performance.

  • Project
    johannfaouzi/pyts

    A Python package for time series classification that provides tools for preprocessing, transformation, and state-of-the-art classification algorithms.

  • Project
    skforecast/skforecast

    A Python library for time series forecasting that provides a unified API for scikit-learn compatible ML models, statistical methods, and zero-shot foundation models.