Preserve PyTorch autograd in partially wrapped normal PDFs

[FlorianPfaff]

Bug

PartiallyWrappedNormalDistribution.pdf(...) evaluated every wrapped Gaussian image through scipy.stats.multivariate_normal.

When the active backend is PyTorch and query points require gradients, SciPy attempts to convert the tensor to NumPy. This raises:

RuntimeError: Can't call numpy() on Tensor that requires grad.

Even for tensors without gradients, the SciPy boundary discards the active backend's dtype, device, and differentiation semantics before converting the result back.

Fix

Evaluate the Gaussian images through PyRecEst's existing backend-native GaussianDistribution.pdf(...) implementation. The wrapping, truncation order, covariance, and summation logic remain unchanged.

This uses SciPy for the NumPy backend, torch.distributions.MultivariateNormal for PyTorch, and jax.scipy.stats.multivariate_normal for JAX through the established Gaussian implementation.

Regression coverage

Add a PyTorch-backend regression that:

• evaluates a two-dimensional partially wrapped normal at two off-boundary points with requires_grad=True;
• verifies tensor type, output shape, dtype, device, positivity, and finiteness;
• backpropagates through the wrapped density;
• verifies finite, nonzero gradients with respect to the query points.

Validation

• branch is based directly on current main (f196ec41d1d8681e279936db9f1a0a3abfa276e7);
• branch is 2 commits ahead and 0 behind;
• final diff is limited to 4 changed production lines and one focused 45-line regression test;
• GitHub Actions is authoritative for the full backend, lint, packaging, documentation, and integration matrix.

[github-actions]

✅MegaLinter analysis: Success

Descriptor	Linter	Files	Fixed	Errors	Warnings	Elapsed time
✅ COPYPASTE	jscpd	yes		no	no	23.6s
✅ JSON	prettier	7	0	0	0	0.93s
✅ JSON	v8r	7		0	0	3.64s
✅ MARKDOWN	markdownlint	68	0	0	0	1.86s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	0.79s
✅ PYTHON	black	1867	88	0	0	89.55s
✅ PYTHON	isort	1867	142	0	0	2.75s
✅ REPOSITORY	betterleaks	yes		no	no	2.41s
✅ REPOSITORY	checkov	yes		no	no	50.63s
✅ REPOSITORY	gitleaks	yes		no	no	17.77s
✅ REPOSITORY	git_diff	yes		no	no	0.27s
✅ REPOSITORY	secretlint	yes		no	no	81.6s
✅ REPOSITORY	syft	yes		no	no	5.64s
✅ REPOSITORY	trivy-sbom	yes		no	no	4.93s
✅ REPOSITORY	trufflehog	yes		no	no	31.75s
✅ YAML	prettier	11	0	0	0	0.73s
✅ YAML	v8r	11		0	0	12.43s
✅ YAML	yamllint	11		0	0	0.62s

Notices

📣 MegaLinter 9.5.0 is out! Discover the new features and security recommendations in the release announcement. (Skip this info by defining SECURITY_SUGGESTIONS: false)

See detailed reports in MegaLinter artifacts

Your project could benefit from a custom flavor, which would allow you to run only the linters you need, and thus improve runtime performances. (Skip this info by defining FLAVOR_SUGGESTIONS: false)

• Documentation: Custom Flavors
• Command: npx mega-linter-runner@9.6.0 --custom-flavor-setup --custom-flavor-linters PYTHON_BLACK,PYTHON_ISORT,COPYPASTE_JSCPD,JSON_V8R,JSON_PRETTIER,MARKDOWN_MARKDOWNLINT,MARKDOWN_MARKDOWN_TABLE_FORMATTER,REPOSITORY_CHECKOV,REPOSITORY_GIT_DIFF,REPOSITORY_GITLEAKS,REPOSITORY_BETTERLEAKS,REPOSITORY_SECRETLINT,REPOSITORY_SYFT,REPOSITORY_TRIVY_SBOM,REPOSITORY_TRUFFLEHOG,YAML_PRETTIER,YAML_YAMLLINT,YAML_V8R

Show us your support by starring ⭐ the repository