KINN 1.0

KINN 1.0 provides a local Python API and CLI for forward and inverse kinetic inference with JAX neural surrogates. Install the kinnlib distribution; use the kinn import and command.

python -m pip install kinnlib
kinn example --method mle --mode inverse --kind adsorption --output problem.json
kinn validate problem.json
kinn run problem.json --output result.json

Python 3.11 or newer is required. Training runs on your machine.

Inputs specify species, the stoichiometric matrix, surface species and vacant sites, observations, initial conditions, neural-network layers and activations, and training controls. Choose fixed-alpha or MLE adaptive covariance weighting, with forward or inverse inference.

The CLI includes 21 saved fixed-weight reference archives, parameter and trajectory exports, and optional plots (pip install "kinnlib[reproduce]"). Full MLE-paper training reproduction is not claimed by this release. General solves report local error-covariance estimates rather than Bayesian credible intervals.

See the CLI guide, reproduction coverage, and citation metadata.

For a GitHub artifact install, download the wheel below and run python -m pip install ./kinnlib-1.0-py3-none-any.whl. SHA256SUMS lists the artifact checksums.