When tuning performance on a model, I sometimes have some idea of parameters that seem to work well, either from a previous related model, trial-and-error, or some understanding I have of the model. However, when tuning with grbtune, I'm not certain that the set of parameters I have is necessarily ideal.
I would like to be able to give my parameter set to grbtune to be used for the baseline run, but they need not be used for subsequent runs if Gurobi thinks it can come up with something better. Right now, if you specify parameters to be used for the baseline run, they are also used for all subsequent runs, preventing grbtune from discovering potentially better parameter sets.
Please sign in to leave a comment.