Bump to NLPModels 0.21 #327
Conversation
|
@amontoison @dpo @abelsiqueira Not sure I have seen this error before https://github.com/JuliaSmoothOptimizers/CUTEst.jl/actions/runs/9226211936/job/25385471817?pr=327#step:7:162 any idea? |
|
Sorry, I missed this. Is the error occurring because of the NLPModels upgrade or because of an upgrade of CUTEst/SifDecode? |
| @@ -44,6 +43,7 @@ function NLPModels.objcons!( | |||
| ) | |||
| end | |||
| increment!(nlp, :neval_obj) | |||
| increment!(nlp, :neval_cons) | |||
There was a problem hiding this comment.
Well, isn't this the source of the error? You're incrementing nevals_cons even on unconstrained problems.
More generally, this pull request is about bumping the NLPModels version, but it does other things. I suggest removing the other changes so we can isolate the problem (if any).
There was a problem hiding this comment.
This is part of the update to NLPModels 0.21. This was necessary to pass unit tests from NLPModelsTest that come along the new release.
Essentially, the idea is that we need to be consistent when calling cons! (and objcons!) on unconstrained problems.
I opened a new PR that updates link to better isolate the issue with the CI on macOS #328
There was a problem hiding this comment.
But what sense is there to incrementing neval_cons on an unconstrained problem?
There was a problem hiding this comment.
I think the discussion goes back to this: https://github.com/JuliaSmoothOptimizers/NLPModels.jl/blob/bfd4ceafec9ad6f58cbf11a4a88f2e61767ce701/src/nlp/api.jl#L60
If a user is calling cons! on an unconstrained problem, should we increment or not? Currently, we do.
|
@tmigot Is it possible to support NLPModels.jl v0.21 now? |
|
superseded by #404 |
No description provided.