Proposal to translate hyp2f1 for complex numbers into Cython
Hi everyone, I've submitted the PR gh-14256, which is a beginning towards moving the implementation of Gauss's hypergeometric function hyp2f1 for complex numbers from Fortran into Cython. Josh Wilson mentioned at https://github.com/scipy/scipy/pull/8151#issuecomment-845227864 that this would be appreciated, and I'd like to work on this little by little in my spare time. The current Fortran implementation has several outstanding issues gh-4894, gh-7340, gh-8054, and gh-8083. Pull requests gh-8110 and gh-8151 were submitted by GitHub user FormerPhysicist to fix some of the outstanding issues but these have languished for almost four years now due to a lack of reviewers conversant in both numerical computation of special functions and Fortran 77. This PR has a fix for gh-7430. I intend to address other defects in the implementation of hyp2f1 in future PRs. Please see my opening comment in gh-14256 for more information on the contents of this PR and plans for the future. Feel free to join the discussion of this update in gh-14256! Thanks, Albert Albert Steppi III, PhD Scientific Software Developer Laboratory of Systems Pharmacology Harvard Medical School
On Sun, Jun 20, 2021 at 5:06 AM Steppi, Albert < Albert_Steppi@hms.harvard.edu> wrote:
Hi everyone,
I've submitted the PR gh-14256, which is a beginning towards moving the implementation of Gauss's hypergeometric function hyp2f1 for complex numbers from Fortran into Cython. Josh Wilson mentioned at
https://github.com/scipy/scipy/pull/8151#issuecomment-845227864
that this would be appreciated, and I'd like to work on this little by little in my spare time. The current Fortran implementation has several outstanding issues gh-4894, gh-7340, gh-8054, and gh-8083. Pull requests gh-8110 and gh-8151 were submitted by GitHub user FormerPhysicist to fix some of the outstanding issues but these have languished for almost four years now due to a lack of reviewers conversant in both numerical computation of special functions and Fortran 77. This PR has a fix for gh-7430. I intend to address other defects in the implementation of hyp2f1 in future PRs. Please see my opening comment in gh-14256 for more information on the contents of this PR and plans for the future.
This sounds great. hyp2f1 has given us headaches for a long time. Thanks for working on this Albert! Cheers, Ralf
Feel free to join the discussion of this update in gh-14256!
Thanks, Albert
Albert Steppi III, PhD Scientific Software Developer Laboratory of Systems Pharmacology Harvard Medical School _______________________________________________ SciPy-Dev mailing list SciPy-Dev@python.org https://mail.python.org/mailman/listinfo/scipy-dev
Hi, Following up on this. It was determined that gh-14256 contains too much content for a single PR. I've closed that PR and am now breaking it up into several smaller PRs. I've submitted gh-14287 which contains only the tests from gh-14256 and the script in scipy.special._precompute that I'm using to generate test cases with mpmath. I've made what I think are improvements to the tests since closing the previous PR. The line count in this PR is also large, but many of the lines are taken up by test cases. I've made the tests self-validating, so that the reviewers will not need to validate that each of the many test-cases test what is claimed. This new PR does not contain specialized numerical mathematics and would be accessible to a wider class of reviewers than the previous PR. Feel free to join the discussion in gh-14256. Thanks, Albert ________________________________ From: SciPy-Dev <scipy-dev-bounces+albert_steppi=hms.harvard.edu@python.org> on behalf of Ralf Gommers <ralf.gommers@gmail.com> Sent: Sunday, June 20, 2021 8:36 AM To: SciPy Developers List <scipy-dev@python.org> Subject: Re: [SciPy-Dev] Proposal to translate hyp2f1 for complex numbers into Cython On Sun, Jun 20, 2021 at 5:06 AM Steppi, Albert <Albert_Steppi@hms.harvard.edu<mailto:Albert_Steppi@hms.harvard.edu>> wrote: Hi everyone, I've submitted the PR gh-14256, which is a beginning towards moving the implementation of Gauss's hypergeometric function hyp2f1 for complex numbers from Fortran into Cython. Josh Wilson mentioned at https://github.com/scipy/scipy/pull/8151#issuecomment-845227864<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_scipy_scipy_pull_8151-23issuecomment-2D845227864&d=DwMFaQ&c=WO-RGvefibhHBZq3fL85hQ&r=6G-pYXDrRNN7Q79BwRHXuvnsUFnhjXBhWoUaZsjutXQ&m=HtKJE76TYmrXzjM4DJ4G0Axt6WsYfb0g1pKaQ2FfOYs&s=3v6PThX-Bs4OSsxmYL2fJcECuU1krGRSLuF1h9QGQ6k&e=> that this would be appreciated, and I'd like to work on this little by little in my spare time. The current Fortran implementation has several outstanding issues gh-4894, gh-7340, gh-8054, and gh-8083. Pull requests gh-8110 and gh-8151 were submitted by GitHub user FormerPhysicist to fix some of the outstanding issues but these have languished for almost four years now due to a lack of reviewers conversant in both numerical computation of special functions and Fortran 77. This PR has a fix for gh-7430. I intend to address other defects in the implementation of hyp2f1 in future PRs. Please see my opening comment in gh-14256 for more information on the contents of this PR and plans for the future. This sounds great. hyp2f1 has given us headaches for a long time. Thanks for working on this Albert! Cheers, Ralf Feel free to join the discussion of this update in gh-14256! Thanks, Albert Albert Steppi III, PhD Scientific Software Developer Laboratory of Systems Pharmacology Harvard Medical School _______________________________________________ SciPy-Dev mailing list SciPy-Dev@python.org<mailto:SciPy-Dev@python.org> https://mail.python.org/mailman/listinfo/scipy-dev<https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.python.org_mailman_listinfo_scipy-2Ddev&d=DwMFaQ&c=WO-RGvefibhHBZq3fL85hQ&r=6G-pYXDrRNN7Q79BwRHXuvnsUFnhjXBhWoUaZsjutXQ&m=HtKJE76TYmrXzjM4DJ4G0Axt6WsYfb0g1pKaQ2FfOYs&s=Jn4XyXE7z9jNOvj0g1CYb12fI2_QpkzWynRgfsqY74Y&e=>
participants (2)
-
Ralf Gommers -
Steppi, Albert