Objective and Constraint rules for SubModels in PAO - pyomo

Question

My question refers to the PAO package, I hope this is the right place to ask.
I am wondering whether it is possible to create lower-level objectives or constraints in a bilevel pao model using "rule"-functions.
I would like to create a lower-level objective and tried the following:
import pyomo.environ as pe
import pyomo.opt as po
from pao.pyomo import *
M = pe.ConcreteModel()
M.S = pe.RangeSet(1, 10)
M.x = pe.Var(M.S, bounds=(0, None))
M.L = SubModel(fixed=M.x)
M.L.y = pe.Var(M.S, bounds=(0, None))
def ul_obj_rule(M):
"""upper-level objective"""
return(sum(M.x[i] + M.L.y[i] for i in M.S))
M.obj = pe.Objective(rule=ul_obj_rule, sense=pe.minimize)
def ll_obj_rule(M):
return(sum(M.L.y[i] for i in M.S))
M.L.obj = pe.Objective(rule=ll_obj_rule, sense=pe.maximize)
This resulted in an error:
AttributeError: 'SubModel' object has no attribute 'S'
I tried adding a Set identical to M.S to the SubModel:
import pyomo.environ as pe
import pyomo.opt as po
from pao.pyomo import *
M = pe.ConcreteModel()
M.S = pe.RangeSet(1, 10)
M.x = pe.Var(M.S, bounds=(0, None))
M.L = SubModel(fixed=M.x)
M.L.y = pe.Var(M.S, bounds=(0, None))
M.L.S = pe.RangeSet(1, 10)
def ul_obj_rule(M):
"""upper-level objective"""
return(sum(M.x[i] + M.L.y[i] for i in M.S))
M.obj = pe.Objective(rule=ul_obj_rule, sense=pe.minimize)
def ll_obj_rule(M):
return(sum(M.L.y[i] for i in M.S))
M.L.obj = pe.Objective(rule=ll_obj_rule, sense=pe.maximize)
which resulted in:
AttributeError: 'SubModel' object has no attribute 'L'
Creating the same objective but using "expr" instead of "rule" works fine:
import pyomo.environ as pe
import pyomo.opt as po
from pao.pyomo import *
M = pe.ConcreteModel()
M.S = pe.RangeSet(1, 10)
M.x = pe.Var(M.S, bounds=(0, None))
M.L = SubModel(fixed=M.x)
M.L.y = pe.Var(M.S, bounds=(0, None))
M.obj = pe.Objective(expr=sum(M.x[i] + M.L.y[i] for i in M.S), sense=pe.minimize)
M.L.obj = pe.Objective(expr=sum(M.L.y[i] for i in M.S), sense=pe.maximize)
I encountered the same problem when I tried to initialize constraints. Does anyone know if and how I could set up objective or constraint rules for bilevel models in PAO without encountering the described problems?
Thank you very much in advance!
Edit: solved it using M.model(). This script works:
import pyomo.environ as pe
import pyomo.opt as po
from pao.pyomo import *
M = pe.ConcreteModel()
M.S = pe.RangeSet(1, 10)
M.x = pe.Var(M.S, bounds=(0, None))
M.L = SubModel(fixed=M.x)
M.L.y = pe.Var(M.S, bounds=(0, None))
def ul_obj_rule(M):
"""upper-level objective"""
mo = M.model()
return(sum(mo.x[i] + mo.L.y[i] for i in M.S))
M.obj = pe.Objective(rule=ul_obj_rule, sense=pe.minimize)
def ll_obj_rule(M):
mo = M.model()
return(sum(mo.L.y[i] for i in mo.S))
M.L.obj = pe.Objective(rule=ll_obj_rule, sense=pe.maximize)

Answer 1

This is because the 'argument' passed into the function ll_obj_rule when it is called in the last line is M.L (instead of the M that you are expecting.) {And M.L (The submodel) does not have an attribute L.}
A simple fix would be omitting the L in the ll_obj_rule:
def ll_obj_rule(M):
return(sum(M.y[i] for i in M.S))
On an unrelated note, I have been trying to run pao on google colab unsuccessfully, often encountering this error. "ImportError: cannot import name 'SimpleBlock' from 'pyomo.core'." Have you encountered this error before? Any help would be greatly appreciated