LaDyS

LaDyS (Latent Dynamical Systems) provides PyTorch benchmark scaffolding for latent variable models of neural dynamics.

Documentation

CLI

Run a full LaDyS experiment with:

ladys run -d lorenz -m cassm

The public registries include synthetic datasets such as lorenz and chaotic_rnn, plus model entries such as ilqr_vae, lfads, and ndt.

This builds the dataset, model, trainer, evaluation metrics, and a self-contained run folder through the public ladys.Experiment orchestration API.

Model pages are generated from model class docstrings and config defaults:

python scripts/generate_model_docs.py

Images referenced from model docstrings should live under website/docs/assets/. Use links like ![Diagram](assets/model-diagram.png) in docstrings; generated model pages rewrite those links to the correct relative path.

Serve the documentation site locally with:

mkdocs serve