Closed fserra closed 8 years ago
I thought we would evaluate neither functions nor derivates and instead just look at the expression graphs. If there's something we don't understand, we should just fail.
that is the current functionality... I just though it might be nice to also provide support for weird user defined stuff.
You're pretty screwed if you're solving a nonconvex MINLP without expression graphs
sure, but if some crazy user wants to have some :ifelse in their expression graph, he/she/it is completely screwed.
On Mon, Sep 5, 2016 at 5:12 PM, Miles Lubin notifications@github.com wrote:
You're pretty screwed if you're solving a nonconvex MINLP without expression graphs
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/leethargo/SCIP.jl/issues/11#issuecomment-244768683, or mute the thread https://github.com/notifications/unsubscribe-auth/AFNG9t149ekVDlF_0Hy35MWHXffHTWuKks5qnDFBgaJpZM4J0i5F .
A lot of expressions involving ifelse
can be reformulated using binary variables. ifelse
is more relevant for continuous NLP.
well, this could still be useful for convex minlp, right?
No, use Pajarito :)
SCIP
allows for user defined expressions. One could in principle give the evaluations via callbacks (as I guess other solvers do), in case we cannot understand the expression provided to MPB.This is probably future future work