A bug of setting warm-start basis using gurobipy

回答済み

Chengwenjian Wang

• 2023年03月13日 22:21

Dear Gurobi Support Team,

I am reaching out to you today because I have encountered a bug that I am confused about. I was wondering if you could kindly assist me in finding a solution or provide some guidance on how to approach the issue.

I have the following function for adding warm-start basis for LP solving:

def add_warm_start_basis(self,
                         basis: Basis) -> None:
    for i, var in enumerate(self.model.getVars()):
        var.setAttr('VBasis', basis.vbasis[i])
    for j, constr in enumerate(self.model.getConstrs()):
        constr.setAttr('CBasis', basis.cbasis[j])
    self.model.setParam("LPWarmStart", 2)  # Make warm-start basis work when conduct presolve.

And Basis class is defined as:

@dataclass
class Basis:
    """Information of: vbasis and cbasis."""

    vbasis: np.ndarray
    cbasis: np.ndarray

    def __post_init__(self):
        self.vbasis = self.vbasis.astype(int)
        self.cbasis = self.cbasis.astype(int)

So it seems I have made sure vbasis and cbasis in ``basis'' are int (I have also checked using the debug function). However, I got the following error report:

in add_warm_start_basis
    var.setAttr('VBasis', basis.vbasis[i])
  File "src/gurobipy/var.pxi", line 187, in gurobipy.Var.setAttr
  File "src/gurobipy/attrutil.pxi", line 241, in gurobipy.__setattr
AttributeError: Attribute 'VBasis' takes integer value

 

Thank you in advance for your time and consideration!

 

Best regards,

Cheng

関連するエンティティ

• Error
• Attribute
• Warm Start

1

• 

  Riley Clement

  Gurobi Staff

  • 2023年03月19日 08:31
  • 編集済み

  Hi Cheng,

  Numpy uses its own types, eg

  >>> type(np.array([1])[0])
  numpy.int64

  Numpy is not a dependency of gurobipy (unless matrix API is used) and so we cannot check for this type.  You can however convert individual numpy ints to standard ints using .item()

  >>> type(np.array([1])[0].item())
  int

  but the best course of action in your example is to convert the numpy arrays to lists and set all vbasis and cbasis information in one step, rather than using the enumeration approach which is slow

  model.setAttr("VBasis", model.getVars(), basis.vbasis.tolist())
  model.setAttr("CBasis", model.getConstrs(), basis.cbasis.tolist())

  Depending on your application, it may make sense to convert to lists within your dataclass.

  - Riley

  1
• 

  Chengwenjian Wang

  • 2023年03月14日 12:21

  

  

  

  Hi Riley,

  Thank you very much for your detailed explanation. It works now!

  Best regards,

  Cheng

  0

サインインしてコメントを残してください。