Disable Hessians when not provided by JuMP #12
Conversation
src/algorithm.jl
Outdated
| m.vartypes = fill(:Cont, numvar) | ||
|
|
||
| MathProgBase.initialize(d, [:Grad, :Jac, :Hess]) | ||
| d.disable_2ndorder ? MathProgBase.initialize(d, [:Grad, :Jac]) : MathProgBase.initialize(d, [:Grad, :Jac, :Hess]) |
There was a problem hiding this comment.
Use :Hess in MathProgBase.features_available(d) instead of d.disable_2ndorder. The former is the documented API (https://mathprogbasejl.readthedocs.io/en/latest/nlp.html#features_available) and the latter is an internal implementation detail that works only when d is of a certain type. This should resolve the test failures.
There was a problem hiding this comment.
Thanks for the comments. Hope the second version is better. Alternatively, we could just check if :Hess is contained in MathProgBase.features_available(d) and use either [:Grad, :Jac, :Hess] or [:Grad, :Jac] accordingly. In fact, simply setting to MathProgBase.features_available(d) in lines 96 and 380 worked both in the Pavito-provided and my own examples.
mlubin
left a comment
mlubin
left a comment
There was a problem hiding this comment.
The code change is correct and it looks like the AppVeyor failure is a pre-existing condition. It would be nice to have this use case covered by tests so that it doesn't break in the future but I'll leave that up to @ccoffrin and @chriscoey to enforce or not.
|
I don't know anything about appveyor or windows really. I'll pass on updating this until someone reports it is not working on Windows. |
|
thanks @spockoyno! |
4 participants
Multivariate user-defined functions in JuMP do not currently allow Hessians. This breaks integer optimisation with Pavito; see https://discourse.julialang.org/t/integer-optimisation-with-users-multivariate-functions-possible-in-julia/17924. The proposed workaround resolves this issue.