Obtain the set of fractional solutions when I stop the running of a MIP

Ongoing

Marina AM

• June 14, 2024 10:54
• Edited

Hi,

I would like to obtain all fractional solutions that Gurobi is exploring when the MIP optimization finishes due to a WorkLimit.

I have used this callback in python:

def mycallback(m, where):
    if where == GRB.Callback.MIPNODE:
        status = m.cbGet(GRB.Callback.MIPNODE_STATUS)
        if status == GRB.OPTIMAL:
            fractional = m.cbGetNodeRel(m._xnm)
            m._solutions = np.r_[m._solutions, fractional]


xnm = m.addVars(XNM, vtype = GRB.BINARY, name = "xnm")
m._solutions = np.zeros([0])
m._xnm = xnm

m.optimize(mycallback)

But with this callback I get the fractional variables of all the nodes that are being explored and I would like to get only the information of those nodes that are being explored at the moment I finish the MIP.

Related to

• Callbacks
• Branch-and-Cut

0

• 

  David Torres Sanchez

  Gurobi Staff

  • June 14, 2024 10:56
  • Edited

  Hi Marina,

  I guess this will work; you will get the relaxation solution for all the nodes, including the last one explored before hitting the work limit.
  If you keep and over-write \(\texttt{fractional}\) every time, you will end up with the last relaxation available before hitting the work limit.

  I guess you can also check the \(\texttt{WORK}\) callback attribute manually if you want. Something like:

  def my_callback(model, where):
      if where == GRB.Callback.MIPNODE:
          status = m.cbGet(GRB.Callback.MIPNODE_STATUS)
          work = m.cbGet(GRB.Callback.WORK)
          if status == GRB.OPTIMAL and work > myworklimit - smallnumber:
              fractional = m.cbGetNodeRel(m._x)

  Or, even better, pair relaxation solutions with the current work and then you can filter them as you like at the end.

  Cheers, 
  David

  0
• 

  Marina AM

  • June 14, 2024 11:35

  

  

  Hi David,

  Thanks for your response, I have edited the code because I did not add the code lines where I store all the fractional solutions.

  The idea to get fractional solutions is to take advantage of the results obtained by MIP during the WorkLimit specified. If I get relaxed solution I would lose some information.

  Is there any way to get the state of the tree to know which binary variables have been discarded as 0/1 (the leaf nodes that have been disposed) and thus know with certainty what value they will have?

  Cheers,

  Marina

   

  0

Please sign in to leave a comment.