I'm working on a problem using Pyomo with SCIP as the solver.
I've noticed that sometimes the same instances are solved almost instantly, while other times they take significantly longer (with a difference of 1 or 2 orders of magnitude). I want to explore the branching procedure (or any other relevant information) in each of these cases to improve parameter tuning.
How can I access this information using Pyomo as the interface?
I doubt if you installed SCIP with PIP you can really apply its parameters on Pyomo. Instead by installing SCIP with conda you can directly pass its parameters to the solver. E.g. to analyze the branching strategies you can use the following method:
branching rule priority maxdepth maxbddist description
-------------- -------- -------- --------- -----------
relpscost 10000 -1 100.0% reliability branching on pseudo cost values
pscost 2000 -1 100.0% branching on pseudo cost values
inference 1000 -1 100.0% inference history branching
mostinf 100 -1 100.0% most infeasible branching
leastinf 50 -1 100.0% least infeasible branching
distribution 0 -1 100.0% branching rule based on variable influence on cumulative normal distribution of row activities
fullstrong 0 -1 100.0% full strong branching
cloud 0 -1 100.0% branching rule that considers several alternative LP optima
lookahead 0 -1 100.0% full strong branching over multiple levels
multaggr 0 -1 100.0% fullstrong branching on fractional and multi-aggregated variables
allfullstrong -1000 -1 100.0% all variables full strong branching
vanillafullstrong -2000 -1 100.0% vanilla full strong branching
random -100000 -1 100.0% random variable branching
nodereopt -9000000 -1 100.0% branching rule for node reoptimization
Option file:
m.Options = {"limits/time": 300, "branching/fullstrong/priority": 10001}
SolverFactory('scip').solve(m, tee= True, options = m.Options, load_solutions=True).write()
then you can see something like this in the solver log:
reading user parameter file <scip.set>
===========================
limits/time = 300
branching/fullstrong/priority = 10001
If you would like to see the statistics information of the model, you can easily pass its option to the file option as follows:
m.Options = {"limits/time": 10, "branching/relpscost/priority": 10001, "display/statistics": True}
and the part of branching in statistics will show something like this:
Branching Rules : ExecTime SetupTime BranchLP BranchExt BranchPS Cutoffs DomReds Cuts Conss Children
allfullstrong : 0.00 0.00 0 0 0 0 0 0 0 0
cloud : 0.00 0.00 0 0 0 0 0 0 0 0
distribution : 0.00 0.00 0 0 0 0 0 0 0 0
fullstrong : 0.00 0.00 0 0 0 0 0 0 0 0
gomory : 0.00 0.00 0 0 0 0 0 0 0 0
inference : 0.00 0.00 0 0 0 0 0 0 0 0
leastinf : 0.00 0.00 0 0 0 0 0 0 0 0
lookahead : 0.00 0.00 0 0 0 0 0 0 0 0
mostinf : 0.00 0.00 0 0 0 0 0 0 0 0
multaggr : 0.00 0.00 0 0 0 0 0 0 0 0
nodereopt : 0.00 0.00 0 0 0 0 0 0 0 0
pscost : 0.00 0.00 0 0 0 0 0 0 0 0
random : 0.00 0.00 0 0 0 0 0 0 0 0
relpscost : 0.30 0.00 404 0 0 2 16 0 6 798
vanillafullstrong: 0.00 0.00 0 0 0 0 0 0 0 0
display statisticsafter. $\endgroup$