Setting optimality tolerance for Gurobi

Q&A
, , ,
1

Dear all,

I am struggling to figure out how to improve the tolerance of my Gurobi solver.
In the Gurobi docs the parameters “FeasiblityTol” and “OptimalityTol” can be set directly as a parameter, but when trying to set these via the cmdline_options in the solve function, they cannot be found (the solver_io is the default value “lp”)
Any tips or experiences?

All the best,

Mo

2

Good settings for energy system modelling for Gurobi:

Have a look at BarConvTol to control the convergence tolerance of the barrier/interior point algorithm.

1 Like
3

Hello Mo,

you are trying to solve an oemof.solph based model, right?
You can provide cmdline_options in the form of a dict. The Gurobi parames, e.g. FeasibilityTol and OptimalityTol and others can be used here as keys. I recommend to go with reasonable presets, such as the example from Fabian.

Best,
Johannes

4

Quick Update:

Thanks first off to both Johannes and Fabian for your quick replies!
I was trying to solve an oemof.solph model and Johannes suggestion worked, but not to the expected effect, I would’ve expected the parameter to shave off a fair chunk of the runtime, but in total i think the savings were akin to around 2-5%, am I wrong to expect a significant improvement?

All the best,

Mo

5

Do you have experience with adjusting AggFill? What is the tradeoff between computational time and memory usage?

6

My experience is that we were worse off for both memory and time when not setting AggFill=0.