Retrieving Solution to a Model Copy
Kate
first solve, the values for a subset of the variables are fixed to
determine the cost of the current solution using variable bounds.
Then, the variable fixings are removed by resetting the variable
bounds to the appropriate values. To solve the model the second time,
a copy of the model is created. The values of the variables from the
previous solve are set as start values for all variables. At this
point I am able to successfully set start values for the variables
using the originally defined variable objects. After solving the
model, I am unable to retrieve the solution. When trying to find the
value of a variable, the following error is reported:
Gurobi Exception caught: Error code = 10004
Var::get
What is the proper way to retrieve the solution using the defined
variable objects?
I've tried using only the initial model for both solves, but get
errors when trying to set the initial start values. The solution to
the first solve is lost when changing the variable bounds, so it is
necessary to set the initial start values.
Thanks for your help!
Greg Glockner
Kate Abercrombie Lindsey
Ideally, I would not copy the entire Model. However, I haven't been able to successfully solve the model the second time with an initial solution. When I only modify the bounds on the initial Model, the solution found from the first run is not longer stored. I need the solution from the first solving to be used as the initial feasible solution for the second run to help reduce the solve time. However, when I define the start values using the solution from the first solve I get the following error:
Gurobi Exception caught: Error code = 20001
Variable not in model
The error occurs on variable 1707095, but there are 1852429 variables total. I am setting the start values using the following code in c++:
avars= model->getVars();
for (int i = 0; i < model->get(GRB_IntAttr_NumVars)-1; i++){
avars[i].set(GRB_DoubleAttr_Start,avars[i].get(GRB_DoubleAttr_X));
}
Do you have any ideas why this error could be happening?
This is the reason I started using the copy of the model to solve the second time.
Thanks!
Greg Glockner
This seems like an issue with C++ objects. For example, you could be inadvertently referencing GRBVar objects from one model inside the copied model. One thing to check is to write the model file as an MPS file, along with the start values as a MST file. Then read the model file and the MST file, and adjust the bounds as you expected.
> Ideally, I would not copy the entire Model. However, I haven't been able to successfully solve the model the second time with an initial solution. When I only modify the bounds on the initial Model, the solution found from the first run is not longer stored.
Correct, your code needs to save the Start values, either via an additional array of doubles or via a MST file.
> I need the solution from the first solving to be used as the initial feasible solution for the second run to help reduce the solve time. However, when I define the start values using the solution from the first solve I get the following error:
>
> Gurobi Exception caught: Error code = 20001
> Variable not in model
>
> The error occurs on variable 1707095, but there are 1852429 variables total. I am setting the start values using the following code in c++:
>
> avars= model->getVars();
> for (int i = 0; i < model->get(GRB_IntAttr_NumVars)-1; i++){
> avars[i].set(GRB_DoubleAttr_Start,avars[i].get(GRB_DoubleAttr_X));
> }
It's hard to determine precisely from this code fragment, but one of two things may be happening:
1) If this code fragment corresponds to the case when you have 2 copies of the model, then 'avars' is referring to one model - you need an array of GRBVar objects to refer to the second model.
2) If this code fragment corresponds to the case when you are reusing 1 copy of the model, then you should try to use the X values to set the Start values prior to making any modifications to the model (such as changing the bounds).
> Do you have any ideas why this error could be happening?
One more thing may be that you've got an intervening call to the update() method (or forgotten it or something similar).
Kate Abercrombie Lindsey
>2) If this code fragment corresponds to the case when you are reusing 1 copy of the model, then you should try to use the X values to set the Start values prior to making any modifications to the model (such as changing the bounds).
>One more thing may be that you've got an intervening call to the update() method (or forgotten it or something similar).
Greg Glockner
As you've figured, you can't cross-reference objects from one model to the other. The only way to identify specific variables like this is to do the following steps:
1. Assign unique string identifiers to the VarName attribute of each GRBVar
2. Retrieve the full set of GRBVar objects and filter out the ones that you want
There is no built-in 'find' or 'filter' function in the C++ library, but it's not difficult to code this yourself.
Alternately, you might find it easier to save the incumbent solution in a MIP start file (.mst). This could save you a lot of coding.
> The code was in reference to the case when only one model is used. I am using the X values to set the Start values prior to any modifications to the model. I just don't understand the reason for the "Gurobi Exception caught: Error code = 20001; Variable not in model" error when no changes have been made to the model.
This error code indicates you are referencing a GRBVar object that isn't in that model. It could be a valid variable from another model in memory.