I'm trying to create a custom search in Docplex (python) in a CP model.
The model:

mdl = CpoModel()
a = mdl.integer_var(0, 2, name="a")
b = mdl.integer_var(0, 8, name="b")
c = mdl.integer_var(0, 8, name="c")
mdl.add(a < b)
mdl.add(c < b)
mdl.add(a < c)
mdl.add(b != 3)
mdl.add(mdl.minimize(a + b+c))
msol = mdl.solve(LogPeriod=1, LogVerbosity="Verbose")
msol.print_solution()​

In C++ you can use `IlcGoal` to implement custom search steps.
However, from the documentation is not clear how to do the same in python.
So far, I tried to use Callbacks:

class MySolver(CpoCallback):
    def invoke(self, solver, event, sres):
        if len(sres.solution.var_solutions_list) > 0:
            print("search step")
            vars = pd.DataFrame(
                {
                    "domain": [
                        np.sum(
                            [
                                inter[1] - inter[0] + 1
                                if isinstance(inter, tuple)
                                else 1
                                for inter in v.value
                            ]
                        )
                        if isinstance(v.value, tuple)
                        else 0
                        for v in sres.solution.get_all_var_solutions()
                    ],
                    "varName": [
                        v.expr.name for v in sres.solution.get_all_var_solutions()
                    ],
                }
            )
            if len(vars[(vars.domain == vars.domain.max()) & (vars.domain > 0)]) > 0:
                vn = vars[
                    (vars.domain == vars.domain.max()) & (vars.domain > 0)
                ].varName.values[0]
                vlue = sres.solution.var_solutions_dict[vn].value[0][0]
                sres.solution.var_solutions_dict[vn] = vlue
                print(f"setting to {vn} = {vlue}")
            else:
                print('Solution found')
mdl.add_solver_callback(MySolver())​

However, in this way, it looks like the solver does not take into account the assignment made by the callback. In fact, the callback assigns a value only to c (value 1) and then finds the optimal solution directly.
Is there any way to implement a custom search in docplex.cp?

Thanks

------------------------------
Federico Caselli
------------------------------

#DecisionOptimization

Hello,
you cannot write a custom search similar to C++ goals with the Python API of CP Optimizer.
The usual way to guide the search is by defining "search phases" and possibly "var/value choosers", which are available in Python.
If search phases do not fill your needs, can you send a description of your need?

------------------------------
Olivier Lhomme
------------------------------