Set constraint using calculations from single values of MVar

ユーザーの入力を待っています。

Weihang Zhang

2023年05月10日 19:55

• 
• 
• 

Hi

 

I have a selection variable as MVar:

selection = model.addMVar(n, vtype=gp.GRB.BINARY)

and I want to set up a constraint with individual values of selection in a matrix.

A = np.zeros((m, m))
for i in range(n):
    j, k = indices[i]
    A[j][k] = selection[i]

model.addConstr(gp.max_(A@A@A) <= 5)

Indices here simply return the corresponding indices of each selection in the matrix A. However, it seems like I can not directly set MVar to a numpy matrix. Can you please suggest what are the alternatives that I can make this work? 

関連

• Matrix API

0

• 

  Mario Ruthmair

  • Gurobi Staff

  2023年05月10日 21:41

  Hi,

  I am not fully sure what the mathematical constraint is that you want to model.
  Could you please specify it in mathematical notation?

  Best regards,
  Mario

  0
• 

  Weihang Zhang

  2023年05月10日 22:10

  • 
  • 
  • 

  Sure.

  The exact context is a bit hard to describe, but essentially I am trying to modify the adjacency matrix A of a graph. And the constraint in this modification is that the modified graph does not have a connected component of size greater than 5. 

  The adjacency matrix is quite large, hence I could not model the entire A as MVar, and therefore decided to select a subset of cells in A as decision variables. I hope this make sense. 

  0
• 

  Mario Ruthmair

  • Gurobi Staff

  2023年05月15日 11:05

  You created a 1-dimensional MVar containing n variables. If you want to model the edges in your graph by variables, then it has to be a nxn matrix of variables. But since I guess your graph is very sparse, I would recommend to not create variables for the full matrix but only for the relevant edges.

  Also, you assign a variable to some cells in a numpy matrix, while in the other cells there is value 0. This is not possible with Gurobi.

  Additionally, if A would be an MVar, A@A@A would create a cubic formulation, which is also not possible with Gurobi. You might need a different approach to model connected components.

  0

サインインしてコメントを残してください。