New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add abs tol to some CBF and RBF doctests #32905
Comments
Author: Samuel Lelièvre |
New commits:
|
Commit: |
Branch: u/slelievre/32905 |
comment:2
[Oops. Initially only considered two out of four files. Edited.] With Sage 9.5.beta7:
With the proposed branch (commit 859b7ba):
|
New commits:
|
Changed branch from u/slelievre/32905 to public/32905 |
This comment has been minimized.
This comment has been minimized.
comment:4
(Not a review, just some comments.) The results should only depend on the installed version of arb, not on other system details. The policy until now (at Jeroen's insistence long ago, if I remember correctly) has been to doctest the exact output corresponding to the version packaged by sage-the-distribution. With increasing use of system packages, it may be time to change that, but that should be a conscious decision. I don't see the point of things like |
This comment has been minimized.
This comment has been minimized.
comment:5
Thanks for clarifying that results only vary The policy you mention does not seem to have That said, your suggestions make a lot of sense. I prefer using |
comment:6
For the same package versions, answers might be a bit different on different hardware, in particular if machine floating point is involved in any way. But yes, # |
comment:7
My understanding is So for instance in the following output:
the doctest would test that we have something of the form
where each of |
comment:8
Did doctests from #32869 and #32851 use In comment:2 above, the tests were run using To adapt CBF and RBF doctests with
But beyond that I don't know the range of Checking the outputs of each CBF and RBF doctest One way might be to use the exact output Related: ticket #32211 will upgrade |
comment:9
Replying to @dimpase:
In principle yes, but I am not aware of any example of this situation with arb. (This is frequent with libm operations, less so with basic IEEE-754 correctly rounded operations, and I suspect small variations in the computation steps for which arb uses hardware FP typically don't affect the final result at all.) |
comment:10
Replying to @slel:
Yes.
I don't think the precise versions matter much for most doctests: typically all that happens with new versions is that some error bounds change a little bit but remain of the same order of magnitude. |
comment:11
On the same machine as in comment:2,
passes all tests in the four affected files I'll prepare a new branch with tolerance indications. |
comment:13
Here is enough to let all CBF and RBF doctests pass, Replacing Getting the minimal fix from this branch New commits:
|
Changed branch from public/32905 to u/slelievre/ticket/32905 |
Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:
|
comment:15
I had left out the changes to |
This comment has been minimized.
This comment has been minimized.
comment:17
Looks good to me. I took the patch for the sagemath package for void linux (wip) which uses arb 2.21.1, and all of those tests that were failing now pass. |
Reviewer: Gonzalo Tornaría |
comment:18
Raising priority to put it in the 9.5 radar, as this is affecting many distro packages. |
comment:19
See e.g. |
Changed branch from u/slelievre/ticket/32905 to |
Across arb versions,
CBF
andRBF
can giveslightly different numerical results.
On an early 2014 MacBook Air, macOS 10.14.6,
many Homebrew packages including arb 2.21.1,
testing Sage 9.5.beta7 with random seed 0
gives failures in
src/sage/functions/gamma.py
-- 1 doctest failedsrc/sage/rings/complex_arb.pyx
-- 3 doctests failedsrc/sage/rings/real_arb.pyx
-- 12 doctests failedsrc/sage/symbolic/function.pyx
-- 1 doctest failedFollowing two related tickets merged in Sage 9.5.beta7,
here we adapt a few doctests to numerical variations
between arb 2.19.0 and arb 2.21.1.
CC: @dimpase @mezzarobba @slel @antonio-rojas
Component: numerical
Keywords: arb, CBF, RBF
Author: Samuel Lelièvre
Branch/Commit:
2ee7b99
Reviewer: Gonzalo Tornaría
Issue created by migration from https://trac.sagemath.org/ticket/32905
The text was updated successfully, but these errors were encountered: