save optimization data for stat !=: Optimal nlp problems

[huckl3b3rry87]

added functionality to save design variables, objective function values and constraint values for a nlp when the stat != :Optimal. This very useful when defining tolerances for the feasibility of the model. This works on the cases tested for both IPOPT and KNITRO. These tools both pass back the data even if the result is not feasible and studying this data can be very useful

[mlubin]

This really calls for a way to query solution status via MathProgBase. CC @ccoffrin

[huckl3b3rry87]

So, I guess that this can be closed because now JuMP passes back the solution even if it is not optimal?

I do have another issue related to this though, when it is solving my screen looks like this:

WARNING: Dual solution not available. Check that the model was properly solved and no integer variables are present.
WARNING: Dual solution not available. Check that the model was properly solved and no integer variables are present.
WARNING: Dual solution not available. Check that the model was properly solved and no integer variables are present.
WARNING: Dual solution not available. Check that the model was properly solved and no integer variables are present.
WARNING: Dual solution not available. Check that the model was properly solved and no integer variables are present.

etc....
Can we suppress these warnings somehow (so my boss does not think something is wrong)? I noticed there has been discussion related to this here #91
and
here #325

[mlubin]

You could just not call getdual if you know there's no dual solution to get.

[huckl3b3rry87]

@mlubin oh, I forgot that I was calling that. Thanks!

[mlubin]

This will be addressed by JuliaOpt/MathProgBase.jl#164 and the corresponding changes to JuMP