Cannot create disjunction of matrix constraints

[bremner]

I'm not sure if this is a documentation bug (i.e. it might just not be possible), or a bug in pyscipopt, but trying to add a disjunction constraint of two matrix constraints reports

AttributeError: 'MatrixConstraint' object has no attribute '_lhs'

I guess even if it is not intended to be supported, letting it fall through to a Python error is not ideal.

To Reproduce

Run python on the following script

from pyscipopt import Model, quicksum
import numpy as np
scip = Model()

n=5
shape=(n,1)
x = scip.addMatrixVar(shape, vtype='C', name='x', ub=8)

A1 = np.random.rand(2,5)
A2 = np.random.rand(2,5)

z2 = np.c_[[0,0]]

c1=scip.addMatrixCons(A1 @ x == z2, enforce=False)
c2=scip.addMatrixCons(A2 @ x == z2, enforce=False)

scip.addConsDisjunction([c1,c2])

Expected behavior

Disjunctive Constraint added to model.
System

• OS: linux
• Version 5.6.0
• SCIP version 9.2.1
• How did you install pyscipopt? from A debian package that I made.

[Joao-Dionisio]

Hey @bremner! I don't see why this shouldn't be possible. Do you think you can create a small PR to add this change?

It should be almost identical to addMatrixConsIndicator. I can help out, if needed :)

Also, do you mind giving details about the debian package you mention?

[kcyang98]

The error occurs because ‘addConsDisjunction()’ only supports expression-based constraints (ExprCons), and does not accept MatrixConstraint objects created by ‘addMatrixCons()’.

When calling:

c1 = scip.addMatrixCons(A1 @ x == z2, enforce=False)
c2 = scip.addMatrixCons(A2 @ x == z2, enforce=False)
scip..addConsDisjunction([c1, c2])

'addMatrixCons()' has already inserted the constraints into the model and returned a MatrixConstraint, which internally represents a set of scalar constraints.
However, 'addConsDisjunction()' expects symbolic expression constraints that have not yet been added to the model.

I think if you need to use disjunctions, the constraints must be passed before they are added to the model, for example:

from pyscipopt import Model, quicksum
import numpy as np
scip = Model()

n=5
shape=(n,1)
x = scip.addMatrixVar(shape, vtype='C', name='x', ub=8)

A1 = np.random.rand(2,5)
A2 = np.random.rand(2,5)

z2 = np.c_[[0,0]]

expr1 = (A1 @ x == z2)  
expr2 = (A2 @ x == z2)

scip.addConsDisjunction([expr1[0, 0], expr2[0, 0]])