Fix absolute value of delta m on trust-region radius #27
Conversation
| @@ -47,7 +47,7 @@ function ratio(nlp :: AbstractNLPModel, f :: Float64, f_trial :: Float64, Δm :: | |||
| pred < 0 || throw(TrustRegionException(@sprintf("Nonnegative predicted reduction: pred = %8.1e", pred))) | |||
|
|
|||
| ared = f_trial - f + max(1.0, absf) * 10.0 * ϵ | |||
| if (Δm < 1.0e+4 * ϵ) || (abs(ared) < 1.0e+4 * ϵ * absf) | |||
| if (abs(Δm) < 1.0e+4 * ϵ) || (abs(ared) < 1.0e+4 * ϵ * absf) | |||
There was a problem hiding this comment.
I thought this was trying to check for small delta m, because delta m is negative.
|
Why are we getting a build error now? |
|
I think |
|
Confirmed. |
|
In what version? I'm running 0.5.0 and it throws an error. |
|
That's the strangest thing. On my versions, it also throws, but the Travis version doesn't. |
|
Travis uses 0.4.7 and 0.6. At least 0.4.7 should throw and error I think. It used to. |
|
Tested it. For "NLPModels" and "BenchmarkProfiles" it returned nothing. For the other packages it threw an error. |
|
Okay, so I think I understand it now. Packages that are on METADATA return nothing, and packages not on METADATA throw an error. Packages that are installed without METADATA return |
4046a06 to
39987ec
Compare
|
Makes sense. |
|
Linux-0.6 is failing with a unresolved issue: JuliaLang/julia#18871, but everything else works. |
|
Why is it ok on osx? |
|
Really no idea. Nobody picked it up yet. |
|
Maybe after JuliaLang/julia#18997, it'll work. |
|
Thanks! |
No description provided.