[Harsh28-raj]

Description

What is the purpose of this pull request?

This pull request:

• adds @stdlib/ml/base/loss/float32/squared-epsilon-insensitive-gradient, a single-precision port of @stdlib/ml/base/loss/float64/squared-epsilon-insensitive-gradient, with JavaScript and C implementations.

Related Issues

Does this pull request have any related issues?

This pull request has the following related issues:

• [RFC]: add native float32 support for ML loss functions and activation primitives #16124

Reference Packages

Did the content of this pull request follow one or more existing packages? Answering is optional.

This pull request followed these packages:

• ml/base/loss/float64/squared-epsilon-insensitive-gradient
• ml/base/loss/float32/epsilon-insensitive-gradient (feat: add ml/base/loss/float32/epsilon-insensitive-gradient #16317)

What carried over from the reference(s)?

• Package structure, naming, documentation, examples, benchmark layout, and test shape from the float64 package.
• Float32 conventions from feat: add ml/base/loss/float32/epsilon-insensitive-gradient #16317: f32 rounding of inputs and intermediate results in JS, float types and f-suffixed literals in C, isnanf, STDLIB_CONSTANT_FLOAT32_NAN, the FFFF_F N-API macro, a 1 ULP tolerance in tests, and the additional NaN tests for y and p.

What differed, or what did you have to work out yourself?

• In JS, y - p, z ± e, and the final product are each rounded to single precision to match the C kernel's float arithmetic.
• The -76.8 example is no longer exact in single precision, so it is documented as ~-76.8.
• runner.jl was updated to compute in Float32. Julia was not available locally, so the fixture JSON files were generated by a Node.js script that mirrors the updated runner.jl, applying Math.fround after every operation.

Questions

Any questions for reviewers of this pull request?

• The tiny-value fixtures (same input ranges as the float64 package) produce single-precision subnormal results (as small as ~2.7e-41). Should I keep these ranges, or narrow them to stay in the normal range?
• Happy to regenerate the fixtures with Julia if preferred.

Other

Any other information relevant to this pull request? This may include screenshots, references, and/or implementation notes.

• JavaScript tests pass locally (2021 assertions). Native tests and the C kernel were not compiled locally (no C toolchain available), so I am relying on CI for those.

Checklist

Please ensure the following tasks are completed before submitting this pull request.

• Read, understood, and followed the contributing guidelines.

AI Assistance

When authoring the changes proposed in this PR, did you use any kind of AI assistance?

• Yes
• No

If you answered "yes" above, how did you use AI assistance?

• Code generation (e.g., when writing an implementation or fixing a bug)
• Test/benchmark generation
• Documentation (including examples)
• Research and understanding

Disclosure

If you answered "yes" to using AI assistance, please provide a short disclosure indicating how you used AI assistance. This helps reviewers determine how much scrutiny to apply when reviewing your contribution. Example disclosures: "This PR was written primarily by Claude Code." or "I consulted ChatGPT to understand the codebase, but the proposed changes were fully authored manually by myself.".

This PR was written primarily by Claude Code, by porting the existing float64 package and following the float32 conventions in #16317. I reviewed the changes and ran the JavaScript tests locally.

---

@stdlib-js/reviewers

[stdlib-bot]

👋 Hi there! 👋

And thank you for opening your first pull request! We will review it shortly. 🏃 💨

Getting Started

• Please read our contributing guidelines if you haven't already.
• For development guidance, refer to the development guide.

Next Steps

1. A project maintainer will approve GitHub Actions workflows for your PR.
2. All CI checks must pass before your submission can be fully reviewed.
3. You'll need to address any failures in linting or unit tests.

Running Tests Locally

You can use make to run any of the CI commands locally from the root directory of the stdlib repository:

# Run tests for all packages in the math namespace:
make test TESTS_FILTER=".*/@stdlib/math/.*"

# Run benchmarks for a specific package:
make benchmark BENCHMARKS_FILTER=".*/@stdlib/math/base/special/sin/.*"

If you haven't heard back from us within two weeks, please ping us by tagging the "reviewers" team in a comment on this PR.

If you have any further questions while waiting for a response, please join our Zulip community to chat with project maintainers and other community members.

We appreciate your contribution!

Documentation Links

• Contributing Guidelines
• Development Guide
• Zulip Chat
• make rules for running examples
• make rules for running unit tests
• make rules for running benchmarks

[AryanSharma48]

Left some initial suggestions.

[Harsh28-raj]

Thanks @AryanSharma48! Applied all four suggestions in f70442d. For the docstring p example, I kept the original range( 10.0, stop=0.0, ... ) and only wrapped it in Float32.( ... ), matching #16317, since I think the 0.0 → 50.0 in the suggestion was a copy slip. Let me know if you meant otherwise.