Data type problems combining DataFrames and CPLEXDirect

[Miguel897]

Versions: Pyomo: 5.7.3 Python: 3.7.8

I'm starting to use the interface CPLEXDirect to avoid temporary files and reducing processing time in small problems. I usually introduce parameters in ConcreteModels as DataFrames. I solve my model like this:

import pyomo.environ as pe
opt = pe.SolverFactory('cplex_direct')
model = create_toy_model()
results = opt.solve(model)

When I create my model in this way it works fine:

def create_toy_model():
    model = pe.ConcreteModel()
    model.x1 = pe.Var(within=pe.NonNegativeReals)
    model.x2 = pe.Var(within=pe.NonNegativeReals)
    model.obj = pe.Objective(expr=2 * model.x1 + 2 * model.x2, sense=pe.maximize)
    model.con1 = pe.Constraint(expr=model.x1 + 2 * model.x2 >= 2)
    model.con2 = pe.Constraint(expr=model.x1 + 2 * model.x2 <= 4)

    return model

However when using DataFrame in the model:

def create_toy_model():
    
    import pandas as pd
    A = pd.DataFrame({0: [1, 1], 1: [2, 2]})
    B = pd.DataFrame({0: [2, 4]})
    C = pd.DataFrame({0: [2], 1: [2]})

    model = pe.ConcreteModel()
    model.x1 = pe.Var(within=pe.NonNegativeReals)
    model.x2 = pe.Var(within=pe.NonNegativeReals)
    model.obj = pe.Objective(expr=model.x1 * C.at[0, 0] + model.x2 * C.at[0, 1], sense=pe.maximize)
    model.con1 = pe.Constraint(expr=model.x1 * A.at[0, 0] + model.x2 * A.at[0, 1] >= B.at[0, 0])
    model.con2 = pe.Constraint(expr=model.x1 * A.at[1, 0] + model.x2 * A.at[1, 1] <= B.at[1, 0])

    return model

I get an error:

TypeError: non-float value in input sequence (2)

This error disappears when defining the df with dtype='float16'. The issue is related with the distinction between data types int and numpy.int.

[michaelbynum]

I agree that this is an issue. As a temporary work-around, you could also wrap constants in float - e.g., float(C.at[0, 0])

[blnicho]

@Miguel897 I think this is related to (or even a duplicate of) #31. See that issue for a more thorough explanation of the problem. I second @michaelbynum's work-around suggestion.