Repository navigation
SPD Learn 0.2.1 is a patch release that fixes licensing and attribution, adds tests and a CI job for the oldest supported dependencies, and prepares the package for the JMLR Machine Learning Open Source Software submission.
Fixed
- Corrected the attribution of the GREEN model and its wavelet layer: both are
re-implementations of GREEN (Paillard, Hipp and Engemann, 2025;
https://github.com/Roche/neuro-green) and are distributed under BSD-3-Clause
like the rest of the package. - The integration test job now runs on pushes to
main(it was gated on a
masterbranch that does not exist).
Tests and continuous integration
- New tests for the logging utilities, the initialization functions, the
manifold parametrizations, the Riemannian PGD projection and the
Log-Euclidean exponential and logarithmic maps, which had no or partial
coverage. - New CI job that installs the lowest dependency versions allowed by
pyproject.tomlon Python 3.11 and runs the test suite.
Packaging
- Declare the license as
BSD-3-Clausein the package metadata (PEP 639),
shipLICENSE.txtas a license file, and add the project URLs. - State the minimum versions of the runtime dependencies:
torch>=2.3,
einops>=0.6,numpy>=2.0. pyRiemann 0.12 declaresnumpy>=1.25but uses
ndarray.mT, which requires NumPy 2.0, and PyTorch supports NumPy 2 from
version 2.3. - Add an SPDX license header to every source file, including tests, examples
and documentation scripts. - Add
CITATION.cffand update the citation in the README and the
documentation (title, full author list, software version). - Remove a duplicated logo image from the documentation sources.
- Keep version-control files, unfinished examples and website verification
files out of source archives (.gitattributesexport-ignore).
Full Changelog: 0.2.0...0.2.1