Python matrix api with non-writable numpy float64 array error: buffer source array is read-only

Answered

Sean Ma

May 09, 2025 01:21

• 
• 
• 

Code to duplicate the error (float32 type is OK - seems the api trying to do calculation in place for the float64 type):

import numpy as np
import gurobipy as gp
model = gp.Model('gurobi_test')
v = model.addMVar(2, vtype=gp.GRB.CONTINUOUS, lb=0.0, ub=[1,2])
c = np.array([1,2], dtype=np.float64)
c.flags.writeable = False
model.setObjective((c * v).sum(), gp.GRB.MINIMIZE)
Error:
ValueError   Traceback (most recent call last) 
Cell In[121], line 7 5 c = np.array([1,2], dtype=np.float64) 
6 c.flags.writeable = False ----> 7 model.setObjective((c * v).sum(), GRB.MINIMIZE) 
File src\\gurobipy\\mvar.pxi:406, in gurobipy._matrixapi.MVar.__rmul__() 
File src\\gurobipy\\mlinexpr.pxi:1821, in gurobipy._matrixapi.MLinExpr.__rmul__() 
File src\\gurobipy\\mlinexpr.pxi:1640, in gurobipy._matrixapi.MLinExpr.__mul__() 
File src\\gurobipy\\mlinexpr.pxi:1743, in gurobipy._matrixapi.MLinExpr._mul_compact_ndarray() 
File src\\gurobipy\\mlinexpr.pxi:1484, in gurobipy._matrixapi.MLinExpr.__imul__() 
File src\\gurobipy\\mlinexpr.pxi:1554, in gurobipy._matrixapi.MLinExpr._imul_compact_ndarray() 
File <stringsource>:663, in View.MemoryView.memoryview_cwrapper() 
File <stringsource>:353, in View.MemoryView.memoryview.__cinit__() 
ValueError: buffer source array is read-only

Related to

• Pandas

0

• 

  Riley Clement

  • Gurobi Staff

  May 09, 2025 02:30

  Hi Sean, 
  
  I'll ask our dev team to take a look.  Which version of gurobipy are you using?

  Thanks,
  Riley

  0
• 

  Sean Ma

  May 09, 2025 03:11

  • 
  • 
  • 

  Thanks Riley. Gurobipy version: '12.0.0'. 

  0
• 

  Riley Clement

  • Gurobi Staff

  May 09, 2025 06:12

  Hi Sean,

  I can confirm this is unintended behavior and we've added an item to our development backlog to patch it in a future release.

  We don't do the calculation in place, and we don't modify the array, but we haven't promised the compiler that we won't modify it either so we get an error when the float64 array has flags.writeable=False.

  Are you running into this problem in production, or was this just to demonstrate the bug for us?

  If you need to keep flags.writeable=True and consequently need a workaround there's a couple of options.  For this specific example, (c*v).sum() is the dot product and is equivalent to c@v, and this latter form does not produce the error (and is faster to build).

  Otherwise you can use the array's copy function to produce a temporary copy which by default will have flags.writeable=True, i.e.

  model.setObjective((c.copy() * v).sum(), gp.GRB.MINIMIZE)

  Thanks,
  Riley

   

  0
• 

  Sean Ma

  May 11, 2025 21:07

  • 
  • 
  • 

  Thanks Riley for the quick response. 

  Some data manipulation using pandas / polars DataFrame will lead to flags.writable=False. 

  I am currently using the array's copy as a workaround. 

  0
• 

  Riley Clement

  • Gurobi Staff

  July 16, 2025 00:00

  Hi Sean,

  This issue has been fixed in v12.0.3 which has just been released.

  https://docs.gurobi.com/projects/optimizer/en/current/reference/releasenotes/fixedbugs.html#fixed-issues-in-gurobi-12-0-3

  Thanks again for reporting!

  Riley

  0
• 

  Sean Ma

  July 16, 2025 22:12

  • 
  • 
  • 

  Thanks Riley.

  0

Please sign in to leave a comment.