Can Yalmip set incomplete initial guesses as MIP start for Cplex?

[Xinwei Shen]

Note that in manual of CPLEX it said "A MIP start may be a feasible solution of the model, but it need not be; it may even be infeasible or incomplete (P279, Starting from a solution: MIP starts)." 

However, in https://yalmip.github.io/Initials you mentioned "In some cases, we might want to initialize only a partial number of variables. At the moment, only gurobi supports this".

[Johan Löfberg]

Ok, good to know. I'll add it to the feature requests

[Xinwei Shen]

After some numerical tests, I believe it already worked in current YALMIP, since the calculation time differs if different initial guesses are provided for CPLEX, even though the "ops.usex0" is reset as 0 after the results came out. Is this possible?

[Johan Löfberg]

what will happen is that non-initialized variables are set to 0, i.e. YALMIP does not communicate that they are un-initialized, as it does in the call to gurobi

the options structure is never resetting your options structure 

[Xinwei Shen]

OK I C. Thank you Prof.

1. A following question is that, if I want to provide a complete feasible solution for CPLEX as initial guess, and somehow I only know some feasible solutions for some integer variables, do I need to calculate other continious variables with those known variables fixed? What are your suggestions?

2. Besides, is initial guess also effective if Benders Decompostion is applied in CPLEX (since V12.7 BD is embedded in CPLEX)? 

[Johan Löfberg]

cplex perhaps tries to fix infeasible initial solutions, but if you can supply from start that might be better, don't know

and don't know anything about benders