Opened 6 years ago

Closed 6 years ago

#21778 closed defect (fixed)

correct refinement in cvxopt sdp backend

Reported by: Dima Pasechnik Owned by:
Priority: major Milestone: sage-7.5
Component: numerical Keywords:
Cc: Matthias Köppe, François Bissey Merged in:
Authors: Dima Pasechnik Reviewers: Matthias Koeppe, François Bissey
Report Upstream: N/A Work issues:
Branch: 15690f5 (Commits, GitHub, GitLab) Commit: 15690f53cadb93290e5c8db5ce9d063d20389032
Dependencies: Stopgaps:

Status badges

Description

on #20506 a weird numerical instability was noticed. It turned out to be a wrongly set solver parameter.

Change History (14)

comment:1 Changed 6 years ago by Dima Pasechnik

Authors: Dima Pasechnik
Branch: u/dimpase/t21778
Cc: Matthias Köppe added
Commit: 4b6e69bf85ae8acc7892c1fde4620a0b5ee9e5e3
Status: newneeds_review

New commits:

f051884Revert "removing numerically unstable test - will be on another ticket"
4b6e69bchanged refinement to 1, adjusted tolerances

comment:2 Changed 6 years ago by Matthias Köppe

On Mac OS X:

sage -t --warn-long 77.7 src/sage/numerical/sdp.pyx
**********************************************************************
File "src/sage/numerical/sdp.pyx", line 97, in sage.numerical.sdp
Failed example:
    p.dual_variable(1)  # rel tol 1e-05
Expected:
    [ 85555.0 -85555.0]
    [-85555.0  85555.0]
Got:
    [ 85527.39489238134 -85527.39489234997]
    [-85527.39489234997  85527.39489238121]
Tolerance exceeded in 4 of 4:
    85555.0 vs 85527.39489238134, tolerance 3e-04 > 1e-05
    -85555.0 vs -85527.39489234997, tolerance 3e-04 > 1e-05
    -85555.0 vs -85527.39489234997, tolerance 3e-04 > 1e-05
    85555.0 vs 85527.39489238121, tolerance 3e-04 > 1e-05
**********************************************************************

comment:3 Changed 6 years ago by git

Commit: 4b6e69bf85ae8acc7892c1fde4620a0b5ee9e5e32dc7460857a4ff8f22cb7ea93e72509a3c198a0e

Branch pushed to git repo; I updated commit sha1. New commits:

2dc7460more tolerance

comment:4 Changed 6 years ago by Dima Pasechnik

OK, I decreased it...

comment:5 Changed 6 years ago by Matthias Köppe

Cc: François Bissey added
Reviewers: Matthias Koeppe

Works on Mac OS X now.

Cc'ing fbissey, who did some testing of this or related testcases on #20506.

comment:6 Changed 6 years ago by Dima Pasechnik

ping!

comment:7 Changed 6 years ago by François Bissey

Sorry, was that the one where I had to raise the tolerance when upgrading glpk?

comment:8 Changed 6 years ago by Dima Pasechnik

No, it is a different issue, with cvxopt interface that was broken and fixed by this patch.

Last edited 6 years ago by Dima Pasechnik (previous) (diff)

comment:9 Changed 6 years ago by François Bissey

OK now that I have the refresher I know what to do. to review this.

comment:10 Changed 6 years ago by François Bissey

Reviewers: Matthias KoeppeMatthias Koeppe, François Bissey
Status: needs_reviewneeds_work

OK the example of #20506 does work now. But I have a tolerance problem

sage -t --long /usr/lib64/python2.7/site-packages/sage/numerical/sdp.pyx
**********************************************************************
File "/usr/lib64/python2.7/site-packages/sage/numerical/sdp.pyx", line 97, in sage.numerical.sdp
Failed example:
    p.dual_variable(1)  # rel tol 1e-04
Expected:
    [ 85555.0 -85555.0]
    [-85555.0  85555.0]
Got:
    [ 85527.39489238134 -85527.39489234997]
    [-85527.39489234997  85527.39489238121]
Tolerance exceeded in 4 of 4:
    85555.0 vs 85527.39489238134, tolerance 3e-04 > 1e-04
    -85555.0 vs -85527.39489234997, tolerance 3e-04 > 1e-04
    -85555.0 vs -85527.39489234997, tolerance 3e-04 > 1e-04
    85555.0 vs 85527.39489238121, tolerance 3e-04 > 1e-04
**********************************************************************

comment:11 Changed 6 years ago by git

Commit: 2dc7460857a4ff8f22cb7ea93e72509a3c198a0e15690f53cadb93290e5c8db5ce9d063d20389032

Branch pushed to git repo; I updated commit sha1. New commits:

6640822Merge branch 'u/dimpase/t21778' of trac.sagemath.org:sage into tolfix
15690f5increased toleranlce 10-fold in the failing test

comment:12 Changed 6 years ago by Dima Pasechnik

Status: needs_workneeds_review

OK, fixed.

comment:13 Changed 6 years ago by François Bissey

Status: needs_reviewpositive_review

Should be all good now.

comment:14 Changed 6 years ago by Volker Braun

Branch: u/dimpase/t2177815690f53cadb93290e5c8db5ce9d063d20389032
Resolution: fixed
Status: positive_reviewclosed
Note: See TracTickets for help on using tickets.