Preserve PyTorch autograd in Mardia-Sutton PDFs

[FlorianPfaff]

Bug

MardiaSuttonDistribution.pdf(...) evaluated its conditional Gaussian factor with scipy.stats.norm.pdf(...).

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 needlessly moves the Gaussian calculation outside the active backend.

Fix

Evaluate the same univariate Gaussian density directly with PyRecEst backend primitives:

exp(-0.5 * ((linear - muc) / sigmac) ** 2) / (sqrt(2 * pi) * sigmac)

The von-Mises factor and distribution parameters are unchanged. The PDF now preserves backend tensor type, dtype, device, and autograd history.

Regression coverage

Add a PyTorch-backend regression that:

• evaluates two off-mode query points with requires_grad=True;
• verifies finite, positive tensor-valued densities with matching dtype and device;
• backpropagates through the PDF;
• verifies finite, nonzero gradients with respect to the query points.

Validation

• reproduced the old SciPy conversion failure with PyTorch 2.10;
• compared the backend-native formula against SciPy for the regression parameters with zero observed absolute difference;
• verified finite, nonzero gradients for both circular and linear coordinates;
• branch comparison: 2 commits ahead and 0 behind main;
• final diff is limited to 3 added / 1 removed production lines and one focused regression file.

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	27.87s
✅ JSON	prettier	7	0	0	0	1.13s
✅ JSON	v8r	7		0	0	5.35s
✅ MARKDOWN	markdownlint	68	0	0	0	2.52s
✅ MARKDOWN	markdown-table-formatter	68	0	0	0	1.13s
✅ PYTHON	black	1866	87	0	0	102.03s
✅ PYTHON	isort	1866	140	0	0	2.82s
✅ REPOSITORY	betterleaks	yes		no	no	2.68s
✅ REPOSITORY	checkov	yes		no	no	55.02s
✅ REPOSITORY	gitleaks	yes		no	no	18.09s
✅ REPOSITORY	git_diff	yes		no	no	0.34s
✅ REPOSITORY	secretlint	yes		no	no	88.43s
✅ REPOSITORY	syft	yes		no	no	6.22s
✅ REPOSITORY	trivy-sbom	yes		no	no	6.17s
✅ REPOSITORY	trufflehog	yes		no	no	33.7s
✅ YAML	prettier	11	0	0	0	1.05s
✅ YAML	v8r	11		0	0	12.56s
✅ YAML	yamllint	11		0	0	0.65s

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