[avgvi]

Summary

The four retraction docstrings in riemannian_optimizers/retractions/stiefel.py and the ObliqueSteepestSGD docstring are plain """ strings that contain LaTeX in $...$. Python reads the backslashes as escape sequences, which causes three problems:

• \S, \e and \| are invalid escapes. Compiling stiefel.py prints three SyntaxWarnings, for example on the first import from a source checkout or when examples/nanogpt/train.py starts.
• \t and \f are valid escapes, so \text turns into a TAB followed by ext, \to into a TAB followed by o, and \frac into a form feed followed by rac. help() and IDE tooltips show the corrupted text.
• The API reference builds these docstrings as reST through autodoc, and reST has no $...$ math. In a local Sphinx build of docs/, the ObliqueSteepestSGD entry shows $l_1 <TAB>o <TAB>ext{RMS}$ as plain text. Fixing the escapes alone would still render it as $l_1 to text{RMS}$.

This PR makes those docstrings raw strings and writes the math with :math: roles, as the rest of the package does (for example procrustes_step.py:34). Because the strings are now raw, step\\_size becomes step\_size. No code changes.

Before and after

On main:

$ python -W always -c "import emerging_optimizers.riemannian_optimizers.retractions.stiefel"
stiefel.py:68: SyntaxWarning: invalid escape sequence '\S'
stiefel.py:92: SyntaxWarning: invalid escape sequence '\e'
stiefel.py:134: SyntaxWarning: invalid escape sequence '\|'

>>> stiefel.cayley_retraction.__doc__      # excerpts
'$(I - \x0crac{\\eta}{2} A)'               # \f (form feed) + "rac"
>>> ObliqueSteepestSGD.__doc__
' $l_1 \to \text'                          # both \t are real TAB characters

$ ruff check --select W605 emerging_optimizers
Found 11 errors.

Rendered apidocs/riemannian-optimizers.html on main:

Oracle (LMO) with respect to the $l_1       o       ext{RMS}$ norm constraint.

With this change:

$ python -W error::SyntaxWarning -c "import emerging_optimizers.riemannian_optimizers.retractions.stiefel"
$ ruff check .
All checks passed!

Oracle (LMO) with respect to the <span class="math">\(l_1 \to \text{RMS}\)</span> norm constraint.

Verification

• tests/test_normalized_optimizer.py: 36 OK
• tests/test_isospectral.py: 17 OK
• tests/test_spel.py: 32 OK
• sphinx-build -b html docs docs/_build/html: no new warnings
• pre-commit run --files <changed files>: all hooks pass

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[greptile-apps]

Confidence Score: 5/5

[Low impact] The PR appears safe to merge.

Summary

Fixes LaTeX in five docstrings without changing optimizer behavior.

• Optimizer docstrings preserve their formulas and show them as math.

_{Reviews (2) · Last reviewed commit: "Drop the W605 lint rule" · Reviewed by Greptile}

[skyw]

Dropping W605 change can make review simpler

[avgvi]

@skyw Hey I dropped W605 change

[skyw]

/ok to test 898505b