JuMP derivatives example

[rgiordan]

Hi Miles,

A while ago I mentioned maybe putting an example of JuMP gradients and Hessians into this repository. I got sidetracked with some stuff at the end of the semester, but here it is at last.

Please let me know if you'd like to make any changes or improvements.

Thanks, Ryan

[mlubin]

Thanks!

Accessing the .col attribute is a bit ugly though, we don't want people doing this. @IainNZ @joehuchette, is it time to add a wrapper? We have ReverseDiffSparse.getplaceindex, maybe just reexport it from JuMP?

[IainNZ]

In this bit:

this_par = m.colVal;

println("x[1] == ", this_par[x[1].col], " == ", getValue(x[1]), " == 1.0")

Does not just using getValue work?

[IainNZ]

But maybe some JuMP.getIndex thing could be nice

Thanks @rgiordan !

[joehuchette]

JuMP.getCol(::Variable)?

[IainNZ]

Might be confused with getting the whole column of A or something like that

[rgiordan]

Sorry for the slow response, I was backpacking. getValue works fine for the value (and as such is maybe not the right place for it in the example), but you'll need the column index to get the derivative and hessian values.

Let me know what you come up with, and I'll update my example!

[dpo]

Just thought I'd mention this (very) preliminary wrapper here: https://github.com/dpo/NLPModels.jl, which makes the user's code substantially simpler and more readable (in my opinion).

[rgiordan]

I agree, that looks better. I'll update this notebook to use the NLPModels library and getLinearIndex once getLinearIndex is merged into a release. Thanks!

[rgiordan]

I guess this is probably out of date now.