Development

Checks

From a source checkout:

python -m pip install ".[dev,plot,table]"
python -m pytest
python -m build

Tests cover Gaussian conditioning, mixture fitting, covariance learning, missing-condition inference, Student-t quadrature, all 12 regression likelihoods, conditional prediction, persistence and plots. Independent regression posterior checks use conjugate Gaussian cases and allow estimated Monte Carlo error. They do not establish general calibration or convergence.

Platform evidence must identify the interpreter, operating system and source revision. The installed-R checks cover selected Gaussian summary cases and are separate from the normal Python test suite. R is not a runtime dependency.

Dependencies and license

NumPy and SciPy are runtime dependencies. Matplotlib and pandas support plotting and tables. Documentation uses MkDocs and MathJax. See LICENSE for GPL-3.0-or-later terms.

Release distributions

The manual Package release workflow defaults to check: it builds a wheel and source archive, checks metadata and archive contents, and retains the artifacts. It does not upload to an index in this mode.

TestPyPI publication uses main. PyPI publication requires a tag matching the package version. Both publishing modes require successful Windows and Linux CI for the same commit and a separately configured Trusted Publisher/environment. The publishing job downloads the build job's artifacts without rebuilding them.