MLinExpr and LinExpr incompatibilities, how to avoid that?

Answered

Daniele Gioia

• March 25, 2022 13:19
• Edited

Hi, I am working with MVars and MLinExpr and I know that those environments are currently not compatible with  LinExpr and Vars. However, Gurobi automatically switches the type of one of my variables from MVar to Var. Specifically, in Python, 

These are three of my variables: 

import gurobipy as grb
zeta = model.addMVar(
    shape=1,
    lb=-grb.GRB.INFINITY,
    ub=grb.GRB.INFINITY,
    name="zeta",
    vtype=grb.GRB.CONTINUOUS
)
Z = model.addMVar(
    shape=100,
    vtype=grb.GRB.CONTINUOUS,
    name="Z"
)
Y = model.addMVar(
    shape=(50,100),
    vtype=grb.GRB.CONTINUOUS,
    name='Y'
)

Assuming two np arrays with the correct dimensions a and b,

 

One of my constraints is

model.addConstrs(
    Z[s] >= a @ X - b @ Y[:, s] - zeta
    for s in range(100)
)

And it raises the error:

GurobiError: Unsupported type (<class 'gurobipy.MLinExpr'>) for LinExpr addition argument

I initially thought that the problem was 'zeta', but, after some checks, the real problem is Z[s]. In particular, although Z is MVar, Z[s] is Var and generates LinExpr rather than MLinExpr. 

 

How can I solve this issue by maintaining the MVars? 

0

• 

  Simon Bowly

  Gurobi Staff

  • March 29, 2022 01:01

  Hi Daniele,

  We currently have a few issues with interactions between Var/LinExpr and MVar/MLinExpr. There were some fixes added to the most recent release (version 9.5.1) which should address your case. Could you please upgrade to version 9.5.1 and additionally rewrite your constraints as below?

  model.addConstrs(
      a @ X - b @ Y[:, s] - zeta - Z[s] <= 0
      for s in range(100)
  )

  Note - I'm assuming here that X is a 1D MVar (similar to Z) and a and b are 1D numpy arrays with compatible shapes for multiplying with X and Y[:, s].

  Best regards,
  Simon

  1
• 

  Daniele Gioia

  • March 29, 2022 15:16

  

  

  

  Thank you Simon, now it works!

  0

Please sign in to leave a comment.