Support Gurobi's computeIIS() and convertToFixed()

[staadecker]

I just shared my progress with the rest of my team and they mentioned that it would be very important to be able to call Gurobi's .computeIIS() method and .convertToFixed(). The former has proven incredibly helpful in many cases and the later is an active part of our current processes. I'm guessing this should be pretty easy to expose in the PyOptInterface API?

[staadecker]

Alternatively, it might be possible to create a generic .run_solver_function(function_name). Like that users could not only do:

• .run_solver_function("computeIIS")
• .run_solver_function("convertToFixed")

But they could also execute other functions like .run_solver_function("optimizeasync") and .run_solver_function("terminate")

[metab0t]

convertToFixed is easy to expose.

computeIIS requires a little more work because I need to implement APIs to query the IIS status for variable (upper/lower bound) and constraints.

[staadecker]

Great! If we can't access the iis status on the variables that's ok since we'd still be able to write out a .ilp file (containing those variables)

[metab0t]

_converttofixed has been added for Gurobi: db70ec6

I use an underscored name to indicate it is an internal function, and there is no similar function in other solvers. You can call it in Pyoframe.

A more general way would be recording the indices and bounds of integer variables and convert them to continuous one, then provide API to convert it back if needed, like what JuMP.jl does.

[metab0t]

For IIS, I propose to add a model.computeIIS() method and add IISUpperBound, IISLowerBound variable attributes and IIS constraint attribute.

@staadecker Is it OK for you?

[staadecker]

Yes!

[metab0t]

@staadecker IIS support is implemented in 68b5a04

Documentation will be added later.

[staadecker]

Let me know when you release the changes in a new version! (Right now I have to compile your library myself and I rather not set up that process in my CI pipeline)