Dynamics Modeling and Analysis with Data (DyMAD)

DyMAD aims to provide a lightweight and user-friendly toolkit for modeling and analyzing dynamical systems using data-driven approaches, plus an agentic interface for users to interact with the package and build their own workflows.

Currently, we have implemented the following features:

  • Data preprocessing pipeline specialized for time series data.

  • Models for different types of dynamical systems, including
    • Latent Dynamics Models

    • Koopman Bilinear Forms

    • The graph version of the above

  • Training utilities, including
    • Neural-ODE-based optimizer

    • Weak-form optimizer

    • Linear preconditioners

    • Hyperparameter tuning tools

  • Spectral analysis based on Koopman theory
    • Eigenvalues and eigenfunctions

    • Spectrum and pseudospectrum

  • Miscellaneous utilities, including
    • Samplers for inputs and initial conditions

    • Visualization tools

Plus the agentic interface for all the above features.

It is still far from complete, see our Roadmap for more details.

The code is hosted on GitHub.

Developers

The package is developed by the APUS Lab at Penn State, directed by Dr. Daning Huang.

The initial development was led by Dr. Yin Yu, whose PhD thesis is on the topic of data-driven modeling of dynamical systems on graphs.

Explore More