Separate access to the dual values of various constraints
AnsweredHello, I have the following model.
from gurobipy import *
model = Model()
I = [1,2,3]
J = [1,2,3]
D = [1,2,3]
# Variables
x = model.addVars(I, J, D, vtype=GRB.BINARY, name="x")
y = model.addVars(J, D, vtype= GRB.CONTINUOUS, name = "y")
# Model
model.setObjective(quicksum(y[j, d] for j in J for d in D), GRB.MAXIMIZE)
for i in I:
model.addLConstr(quicksum(x[i, j, d] for j in J for d in D) <= 1, name = "x")
for j in J:
for d in D:
model.addLConstr(quicksum(x[i, j, d] for i in I) + y[j, d] == 1, name = "y")
model.optimize()
# Relax
relaxed = model.relax()
relaxed.optimize()
# Duals
duals = relaxed.getAttr("Pi", relaxed.getConstrs())
print(duals)
Unfortunately, the code gives me the dual values for both constraints in a list, but I want to save and access them separately for each constraint.
So, for example, save the dual values for constraint "x" and access them with duals1[i] so that I can then access the 3x1 duals of the first constraint by iterating over I and the other 3x3 dual values with duals2[j,d].
-
-
Gurobi Staff
I would consider using addConstrs() for each of the constraint sets. It gives you back a dictionary, e.g. a mapping of i/j/d indices to constraints. Then you can pass that result to getAttr() and you get back a similar dictionary with duals.
ca = model.addConstrs( (quicksum(x[i, j, d] for j in J for d in D) <= 1 for i in I), name = "x")
cb = model.addConstrs( (quicksum(x[i, j, d] for i in I) + y[j, d] == 1 for j in J for d in D), name = "y")
# Relax and solve
duals = relaxed.getAttr("Pi", ca)
print(duals)0 -
-
Ronald van der Velden Thank you for your help. Sadly it does return this error "gurobipy.GurobiError: Constr not in model". How do i fix that?
0 -
-
-
Ah, sorry about that - I should have tested your exact scenario in full and had ignored the fact that calling relax() returns a new model where you cannot query attributes based on constraints from the original model. Unfortunately there is not a clean way to map elements from the original model to their counterpart in the relaxed model. Personally I would relax the model in-place using the second method from this article.
0 -
-
Ronald van der Velden No worries. I tried your approach and initially it didnt work. But updating the model with "model.update()" after assigning names to the constraints did the trick. Thanks
0 -
Please sign in to leave a comment.
Comments
4 comments