API: Should we merge `addCons` and `addMatrixCons`

[Zeroto521]

Should we merge addCons and addMatrixCons? Or should we let addMatrixCons's cons argument accept ExprCons?
Because MatrixExprCons supports .sum(axis=0) now, and the result could be a MatrixExprCons or a scalar.

from pyscipopt import ExprCons, MatrixExprCons, Model

model = Model()

m = 1  # I don't know the specific value, it is from outside. 
n = 1  # as above
x = model.addMatrixVar((m, n), "x", "I")
y = model.addMatrixVar((m, n), "y", "I")

# I want to compare the result of summing row values
model.addMatrixCons(x.sum(axis=0) == y.sum(axis=0))
# But this raises an error, because `x.sum(axis=0)` is `(1, 1)` shape, and it will be changed to a scalar to compare.

# AssertionError                            Traceback (most recent call last)
# Cell In[9], line 1
# ----> 1 model.addMatrixCons(x.sum(axis=0) == y.sum(axis=0))

# File src/pyscipopt/scip.pxi:5774, in pyscipopt.scip.Model.addMatrixCons()

# AssertionError: given constraint is not MatrixExprCons but ExprCons

Based on this case, I needed to add a helper function.
And I thought the answer to the title question was.

from pyscipopt import (
    Constraint,
    ExprCons,
    MatrixConstraint,
    MatrixExprCons,
    Model,
)

def addCons(
    model: Model, cons: ExprCons | MatrixExprCons, **kwargs
) -> Constraint | MatrixConstraint:
    """
    A helper function to add constraints to the model.
    This function handles both ExprCons and MatrixExprCons types.
    """
    if isinstance(cons, ExprCons):
        return model.addCons(cons, **kwargs)
    elif isinstance(cons, MatrixExprCons):
        return model.addMatrixCons(cons, **kwargs)
    raise TypeError(f"Unsupported constraint type: {type(cons)}")

[Zeroto521]

closes to #1025

[Zeroto521]

addMatrixConsIndicator can also receive a scalar via matrix.sum
addConsIndicator and addMatrixConsIndicator have the same problem.

[Joao-Dionisio]

Hey @Zeroto521! I'd rather change getMatrixCons to accept ExprCons, as separating the two modeling approaches makes sense to me.

Do you know how the other modeling languages approach this issue? Thinking of XpressPy and GurobyPy

[Zeroto521]

I prefer to merge addMatrixCons and addCons. Once addMatrixCons could accept ExprCons. addMatrixCons could do all addCons things, but addCons can't do what addMatrixCons things (like matrix). The coverage of addMatrixCons will be greater than addCons.

• gurobi: use one function to handle both variable and matrix variable, Model.addConstr, Model.addGenConstrIndicator

  # From official web
  model.addConstr(x + y <= 2.0, "c1")  # variable 
  model.addConstr(A @ t >= b)  # matrix variable

• xpress: use one function too, problem.addConstraint, problem.addIndicator

  # From official web
  N = 20
  p = xpress.problem()
  x = [p.addVariable() for i in range(N)]
  c = [x[i] <= x[i+1] for i in range(N-1)]
  c2 = x[0] >= x[19]
  p.addConstraint(x[2] == x[4])  # variable 
  p.addConstraint(c, c2)  # list of variables

[Joao-Dionisio]

After discussing with my colleagues, they were against merging the two functions, to make it more explicit what is happening, and to avoid the same function returning different types. Maybe in the future we'll change our mind, but it's how we stand at the moment.

Maybe we can have the best of both worlds, and have your helper function as a recipe. See here for other recipes.