BARON numLoc & local search preprocessing
Posted: Fri Jun 23, 2023 7:07 pm
Hi everyone,
As Im reading more into what BARON does, the solver manual talks about how BARON does a local search in its preprocessing phase. It states that BARON decides the number of local search in preprocessing based on problem and NLP solver characteristics.
First, how do i specify the numLoc setting? Typically, I would type "Option [option]=[setting value]" right before the "solve model" line. But it doesnt seem to accept it and there is nothing in the manual that specifies how to set it up?
Furthermore, how do we look more into the number of initial points BARON looks at? What are those initial points? If we decide to feed the solver an initial solution, how does that affect the subsequent local searches that it does?
Also, can we essentially let BARON do a brute force approach on all the possible solutions? My model is highly nonlinear and nonconvex that even with a simple scenario, it would take hours to solve. It might be even helpful to look at an arbitrarily large number of initial points first before it does more of the complicated steps. Can we leverage parallel processing to look at all these points?
Any help on this is greatly appreciated!
As Im reading more into what BARON does, the solver manual talks about how BARON does a local search in its preprocessing phase. It states that BARON decides the number of local search in preprocessing based on problem and NLP solver characteristics.
First, how do i specify the numLoc setting? Typically, I would type "Option [option]=[setting value]" right before the "solve model" line. But it doesnt seem to accept it and there is nothing in the manual that specifies how to set it up?
Furthermore, how do we look more into the number of initial points BARON looks at? What are those initial points? If we decide to feed the solver an initial solution, how does that affect the subsequent local searches that it does?
Also, can we essentially let BARON do a brute force approach on all the possible solutions? My model is highly nonlinear and nonconvex that even with a simple scenario, it would take hours to solve. It might be even helpful to look at an arbitrarily large number of initial points first before it does more of the complicated steps. Can we leverage parallel processing to look at all these points?
Any help on this is greatly appreciated!