Dynamics Modeling and Analysis with Data (DyMAD)

This is the GitHub site for DyMAD, developed and maintained by the APUS Lab at Penn State.

This README is mainly for developer use. For documentation, refer to readthedocs.org.

Commandline Collection

Installation: During development, test the install by

pip install -e .

so that edits in the source are applied directly.

Tests: To run the tests, stay in the root and

pytest

For specific test script

pytest [filename]

For specific case in specific test script

pytest [filename]::[casename]

Documentation: Generate the document by

sphinx-build -E -b html docs docs/_build/html

Remove -E for incremental build.

If there are API errors such as some files not found, try deleting the auto-generated api folder and try again.

Code Structure

• docs: Documentation based on ReadTheDocs.
• examples: Example cases used in documentation - some of them are still in progress of conversion.
• scripts: Quick cases for checking model performance, mainly used during development.
• src: The source code.
• tests: Tests used by pytest and CI workflow. See the README therein for more details.

TODO notes

• Variable length data (Data class does allow this, but collate and training need to adapt)

• Update the aircraft example

• Update the double pendulum example

• FTLE and CLV for nonlinear latent dynamics

• K-Fold CV

• Demo on KS chaotic

• two-stage implementation

• N4SID

• EnKF, EnKS, IEnKS

• Interface to numpy

• Split DynData into regular and graph versions

• Group transforms for DynData, so that trj_mgr and predict can reuse

• Torch-fy transform classes

• Internal normalization in PIROM

• Symmetry analysis

• Torch-fy Manifold class

• L1/L0 regularized optimization