feat: add math/base/tools/normhermitepolyf

[C0ldSmi1e]

Description

What is the purpose of this pull request?

This pull request:

• adds math/base/tools/normhermitepolyf, which is the single precision equivalent for math/base/tools/normhermitepoly.

Related Issues

Does this pull request have any related issues?

This pull request:

• fixes [RFC]: add math/base/tools/normhermitepolyf #2029

Questions

Any questions for reviewers of this pull request?

No.

Other

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

No.

Checklist

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

• Read, understood, and followed the contributing guidelines.

---

@stdlib-js/reviewers

[C0ldSmi1e]

Test Coverage Report.

[Planeshifter]

Left one comment; otherwise this looks all good to me. Thank you for working on this!

[C0ldSmi1e]

Hi @Planeshifter @kgryte , I've converted float16 to float32 in runner.py.

Initially I used float16 was because, if I borrow the generator from normhermitepoly, or use float32, test cases can be so small that the variables can be Infinity and thus NaN. So I tried to reduce the data scale a bit to avoid this situation. Now I use float32 and change 1.0 to 0.1, which can also pass the test and is more reasonable.

Please review it again when you are convenient, thanks for your time and support!

[kgryte]

@daniel777y Not sure I follow on "test cases can be so small that the variables can be Infinity and thus NaN".

[C0ldSmi1e]

@daniel777y Not sure I follow on "test cases can be so small that the variables can be Infinity and thus NaN".

Hi, @kgryte . The failures are mainly from random n and x, small positive and small negative cases. I also gave my thoughts here.

The process is in the middle of the calculation, some x * y2, i * y3 can exceed the range of float32, and if convert them to float32, they will be Infinity or -Infinity. The next line s1 - s2 can be NaN.

for ( i = n; i > 1; i-- ) {
    s1 = float64ToFloat32( x * y2 );
    s2 = float64ToFloat32( i * y3 );
    y1 = float64ToFloat32( s1 - s2 );
    y3 = y2;
    y2 = y1;
}

What is the best practice to deal with this issue?

[kgryte]

@daniel777y After looking at the code, your reasoning is sound. Fine with how you adjusted things.

[kgryte]

LGTM. Thanks, @daniel777y!

[C0ldSmi1e]

@kgryte Thank you! I learned a lot from it, and look forward to working more on this community!