Innovate

Innovate is a contract-first diffusion modeling library. The current design centers on a stable functional kernel, Arrow interchange, optional backend implementations, and language bindings layered on top of the same core behavior.

The Astro/Starlight site is the active documentation surface. This README is the short front door. Markdown prose is checked in CI with Vale to keep the short-form docs clear and consistent.

What lives where

• src/innovate/ contains the Python kernel, fitters, diagnostics, Arrow interchange, and stability layers.
• bindings/ contains the language bindings and their tests.
• docs/astro-site/ contains the active Astro/Starlight documentation site.
• docs/source/_static/ contains machine-readable release, registry, and migration evidence consumed by the Astro/Starlight site and CI.
• conductor/ contains the track history and archived implementation plans.

Core ideas

• The functional kernel is the primary surface for fitting, predicting, simulating, summarizing, and diagnosing models.
• Arrow interchange is the stable cross-language data contract.
• Optional backends extend the same contract without changing the public API.
• Bindings are thin layers over the kernel, not separate model implementations.

Install

python -m pip install innovate

For optional JAX and Bayesian backends:

python -m pip install "innovate[jax,bayesian]"

For contributor setup from a checkout:

uv sync

Quick start

from innovate import fit_model, predict_model

result = fit_model(data)
forecast = predict_model(result, horizon=12)

Read next

• Documentation landing page
• Tutorials
• Bindings hub
• Architecture principles
• Modernization roadmap
• ADR index