generate variables dynamically (try with the runbenders example)

[hoermann....@gmail.com]

Hi all,

i have got a question. Is it possible to generate Variables dynamically? I am thinking of something in the style of runbenders (https://software.sandia.gov/svn/public/pyomo/pyomo/trunk/examples/pyomo/benders/runbenders)  with its master.py and subproblem.py. Now the benders cut is generated as follows:

    # add the master cut.
    cut = sum((mstr_inst.time_price[t,s,i] * \
                mstr_inst.avail[t])
               for t in mstr_inst.TWOPLUSWEEKS
               for s in mstr_inst.SCEN)
    cut += sum((mstr_inst.bal2_price[p,s,i] * \
                 (-mstr_inst.Inv1[p]))
                for p in mstr_inst.PROD
                for s in mstr_inst.SCEN)
    cut += sum((mstr_inst.sell_lim_price[p,t,s,i] * \
                mstr_inst.market[p,t])
               for p in mstr_inst.PROD
               for t in mstr_inst.TWOPLUSWEEKS
               for s in mstr_inst.SCEN)
    mstr_inst.Cut_Defn.add(mstr_inst.Min_Stage2_Profit <= cut)

I am running a similar script. Now i would like to create a set of new variables in the master problem and use them for the cut:

   # my dream
add in master.py
model.x = Var(model.CUTS) # this set has been empty before and has now one element; in the next iteration it can have two and so on

add in runbenders
   mstr_inst.Cut_Defn.add(mstr_inst.Min_Stage2_Profit <= summation(model.x))

Can something like this work?

Kind regards,
Fabian

[Siirola, John D]

Sure.  I have no clue if the modeling concept makes sense, but you can (easily) implement it in Pyomo:

 

>>> m = ConcreteModel()

>>> m.cuts = Set(initialize=[])

>>> m.x = Var(m.cuts, initialize=0)

>>> m.pprint()

1 Set Declarations

    cuts : Dim=0, Dimen=1, Size=0, Domain=None, Ordered=False, Bounds=(None, None)

        []

 

1 Var Declarations

    x : Size=0, Index=cuts, Domain=Reals

        Key : Lower : Value : Upper : Fixed : Stale

 

2 Declarations: cuts x

 

Now you can add members to the indexing set…

 

>>> m.cuts.add(1)

>>> m.pprint()

1 Set Declarations

    cuts : Dim=0, Dimen=1, Size=1, Domain=None, Ordered=False, Bounds=(None, None)

        [1]

 

1 Var Declarations

    x : Size=0, Index=cuts, Domain=Reals

        Key : Lower : Value : Upper : Fixed : Stale

 

2 Declarations: cuts x

 

…and when you use the “new” variables, “the right thing happens” (i.e., Pyomo will instantiate the “new” variable and initialize it per the original Var declaration)

 

>>> print “new variable = “, m.x[1].value

new variable = 0

>>> m.pprint()

1 Set Declarations

    cuts : Dim=0, Dimen=1, Size=1, Domain=None, Ordered=False, Bounds=(None, None)

        [1]

 

1 Var Declarations

    x : Size=1, Index=cuts, Domain=Reals

        Key : Lower : Value : Upper : Fixed : Stale

          1 :  None :  0 :  None : False :  False

 

2 Declarations: cuts x

 

john

--
You received this message because you are subscribed to the Google Groups "Pyomo Forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pyomo-forum...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.