Skip to first unread message
Ammon Washburn
Aug 29, 2016, 6:55:55 PM8/29/16
to YALMIP
Hello,
I wanted to speed up my code using the optimizer command. I implemented the optimizer command with parameters and checked it against the old (working) code that used optimize and it seemed to do well. They didn't get exactly the same output but it was close enough. My problem is a SOCP and I assumed that because sedumi uses iterative methods then the differences were due to that.
The problem came when I saw that on the page of the more general optimizer (http://users.isy.liu.se/johanl/yalmip/pmwiki.php?n=Blog.Beta-version-of-a-more-general-optimizer) that you said that the non-linearly parameterized feature isn't available in the cone operator.
Was I getting lucky in how close they were or should I expect to see them close but different? I have tried your fix (cone(u,e), u == param * t) mentioned on the page but it seems to mess up my results from old code that I am pretty sure is working. Is that normal as well?
Johan Löfberg
Aug 30, 2016, 3:48:15 AM8/30/16
to YALMIP
Nonlinear paramerization should work in cone.
If not, supply a reproducible example
Johan Löfberg
Aug 30, 2016, 3:59:24 AM8/30/16
to YALMIP
If the difference is larger than, say, 1e-7, (and the solution is unique), you probably have something wrong
Ammon Washburn
Aug 30, 2016, 11:55:11 AM8/30/16
to YALMIP
It does seem to work to me. It was just this sentence on your website made me worried: "Terms in a cone operator can not be nonlinearly parameterized ". However since you said it was fine then I will just assume the website isn't updated.
Johan Löfberg
Aug 30, 2016, 2:59:28 PM8/30/16
to YALMIP
Good to hear. And the differences you see are small enough to be effects of solver tolerances...?
Ammon Washburn
Aug 31, 2016, 2:28:47 PM8/31/16
to YALMIP
The differences are very small to the effects of solver tolerances (scale of 10^-14). Sorry for not saying that before .
Reply all
Reply to author
Forward
0 new messages