[lp.c:8022] ERROR: invalid orthogonality function parameter 's' scip

[leinace1001]

When I try to tune scip hyperparameters related to separating and cutselection, this error throws out some times.

I am wondering which separating or cutselection method may cause this error?

[Joao-Dionisio]

Hey @leinace1001 ! Looking at the scip repo, it seems that your issue is called from SCIProwGetParallelism(), which only accepts 'e' and 'd' as inputs. Do you know which parameter you're providing 's' as an argument to?

[leinace1001]

t the scip repo, it seems that your issue is called from SCIProwGetParallelism(), which only accepts 'e' and 'd' as inputs. Do you know which parameter you're providing 's' as an argument to?

I don't think any parameter is 's', here is my code for searching space:

    cs = ConfigurationSpace()   

    # branching
    cs.add(UniformIntegerHyperparameter("branching/relpscost/priority", -20000, 20000, default_value=-17817))
    cs.add(UniformIntegerHyperparameter("branching/pscost/priority", -20000, 20000, default_value=-13145))
    cs.add(UniformIntegerHyperparameter("branching/inference/priority", 0, 20000, default_value=15684))
    cs.add(UniformIntegerHyperparameter("branching/mostinf/priority", -20000, 20000, default_value=7087))
    cs.add(UniformIntegerHyperparameter("branching/leastinf/priority", -20000, 20000, default_value=-1839))
    cs.add(UniformIntegerHyperparameter("branching/fullstrong/priority", -20000, 5000, default_value=-8152))
    cs.add(UniformIntegerHyperparameter("branching/distribution/priority", -20000, 20000, default_value=-10753))
    cs.add(UniformIntegerHyperparameter("branching/lookahead/priority", -20000, 20000, default_value=5694))
    cs.add(UniformIntegerHyperparameter("branching/multaggr/priority", -20000, 20000, default_value=351))
    cs.add(UniformIntegerHyperparameter("branching/cloud/priority", -20000, 20000, default_value=-2336))
    cs.add(UniformIntegerHyperparameter("branching/allfullstrong/priority", -20000, 5000, default_value=-5189))
    cs.add(UniformIntegerHyperparameter("branching/gomory/priority", -20000, 20000, default_value=-19186))
    # separating
    cs.add(UniformIntegerHyperparameter("separating/closecuts/priority", -536870912,536870911, default_value=1000000))
    cs.add(CategoricalHyperparameter("separating/closecuts/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/rlt/priority", -536870912,536870911, default_value=10))
    cs.add(CategoricalHyperparameter("separating/rlt/freq", [-1, 0, 1, 5, 10, 20], default_value=0))
    cs.add(UniformIntegerHyperparameter("separating/disjunctive/priority", -536870912,536870911, default_value=10))
    cs.add(CategoricalHyperparameter("separating/disjunctive/freq", [-1, 0, 1, 5, 10, 20], default_value=0))
    cs.add(UniformIntegerHyperparameter("separating/gauge/priority", -536870912,536870911, default_value=0))
    cs.add(CategoricalHyperparameter("separating/gauge/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/interminor/priority", -536870912,536870911, default_value=0))
    cs.add(CategoricalHyperparameter("separating/interminor/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/minor/priority", -536870912,536870911, default_value=0))
    cs.add(CategoricalHyperparameter("separating/minor/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/convexproj/priority", -536870912,536870911, default_value=0))
    cs.add(CategoricalHyperparameter("separating/convexproj/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/mixing/priority", -536870912,536870911, default_value=-50))
    cs.add(CategoricalHyperparameter("separating/mixing/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/impliedbounds/priority", -536870912,536870911, default_value=-50))
    cs.add(CategoricalHyperparameter("separating/impliedbounds/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/intobj/priority", -536870912,536870911, default_value=-100))
    cs.add(CategoricalHyperparameter("separating/intobj/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/cgmip/priority", -536870912,536870911, default_value=-1000))
    cs.add(CategoricalHyperparameter("separating/cgmip/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/gomory/priority", -536870912,536870911, default_value=-1000))
    cs.add(CategoricalHyperparameter("separating/gomory/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/strongcg/priority", -536870912,536870911, default_value=-10000))
    cs.add(CategoricalHyperparameter("separating/strongcg/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/gomorymi/priority", -536870912,536870911, default_value=-10000))
    cs.add(CategoricalHyperparameter("separating/gomorymi/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/flowcover/priority", -536870912,536870911, default_value=-100000))
    cs.add(CategoricalHyperparameter("separating/flowcover/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/cmir/priority", -536870912,536870911, default_value=-100000))
    cs.add(CategoricalHyperparameter("separating/cmir/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/knapsackcover/priority", -536870912,536870911, default_value=-100000))
    cs.add(CategoricalHyperparameter("separating/knapsackcover/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/aggregation/priority", -536870912,536870911, default_value=-3000))
    cs.add(CategoricalHyperparameter("separating/aggregation/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/clique/priority", -536870912,536870911, default_value=-5000))
    cs.add(CategoricalHyperparameter("separating/clique/freq", [-1, 0, 1, 5, 10, 20], default_value=0))
    cs.add(UniformIntegerHyperparameter("separating/zerohalf/priority", -536870912,536870911, default_value=-6000))
    cs.add(CategoricalHyperparameter("separating/zerohalf/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/lagromory/priority", -536870912,536870911, default_value=-8000))
    cs.add(CategoricalHyperparameter("separating/lagromory/freq", [-1, 0, 1, 5, 10, 20], default_value=10))
    cs.add(UniformIntegerHyperparameter("separating/mcf/priority", -536870912,536870911, default_value=-10000))
    cs.add(CategoricalHyperparameter("separating/mcf/freq", [-1, 0, 1, 5, 10, 20], default_value=0))
    cs.add(UniformIntegerHyperparameter("separating/eccuts/priority", -536870912,536870911, default_value=-13000))
    cs.add(CategoricalHyperparameter("separating/eccuts/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/oddcycle/priority", -536870912,536870911, default_value=-15000))
    cs.add(CategoricalHyperparameter("separating/oddcycle/freq", [-1, 0, 1, 5, 10, 20], default_value=-1))
    cs.add(UniformIntegerHyperparameter("separating/rapidlearning/priority", -536870912,536870911, default_value=-1200000))
    cs.add(CategoricalHyperparameter("separating/rapidlearning/freq", [-1, 0, 1, 5, 10, 20], default_value=5))
    # cutseletion
    cs.add(UniformIntegerHyperparameter("cutselection/hybrid/priority", -536870912,1073741823, default_value=8000))
    cs.add(UniformIntegerHyperparameter("cutselection/ensemble/priority", -536870912,1073741823, default_value=7000))
    cs.add(UniformIntegerHyperparameter("cutselection/dynamic/priority", -536870912,1073741823, default_value=7000))

[Joao-Dionisio]

Oh, it's actually a SCIP bug! One of your settings calls a function that's not usually called, which has the following line:

cosine = SCIProwGetParallelism(bestcut, cuts[i], 's');

Thank you for the issue! Unfortunately, it has to be patched in SCIP, meaning that you will either have to compile PySCIPOpt with this new SCIP yourself, or wait for a new release (less than a month, I'll say).

And as for which of your settings is calling this function, I't probably the last one. The buggy function is filterWithDynamicParallelism(), which is only called by SCIPselectCutsDynamic(). So, giving a higher priority to dynamic cut selection should increase the likelihood of this function being called.

[Joao-Dionisio]

If you want, you can take a look at 2c79dc4 with the fix.

[Joao-Dionisio]

There are many open issues now and I'm itching all over, so I'm closing this one, as it's essentially solved :) Feel free to reopen if you feel that's not the case!

Also, please let me know if you find a non-default configuration of parameters that ends up being quite good!