nl_writer does not cast Boolean variable values to [0, 1]

[mdealencar]

Summary

When using SCIP as the solver (possibly with other .nl file-based solvers as well), any initialized Boolean variable will make SCIP fail ("ERROR: expected double"). The obvious workaround is to declare the variable as Binary and explicitly cast (False, True) to (0, 1).

Steps to reproduce the issue

# adapted from knapsack-pyomo.py

import pyomo.environ as pyo

v = {'hammer':8, 'wrench':3, 'screwdriver':6, 'towel':11}
w = {'hammer':5, 'wrench':7, 'screwdriver':4, 'towel':3}
limit = 14
items = list(sorted(v.keys()))

# Create model
m = pyo.ConcreteModel()

# Variables
m.x = pyo.Var(items, within=pyo.Boolean, initialize=True)

# Objective
m.value = pyo.Objective(expr=sum(v[i]*m.x[i] for i in items), sense=pyo.maximize)

# Constraint
m.weight = pyo.Constraint(expr=sum(w[i]*m.x[i] for i in items) <= limit)

# Write .nl
m.write('knapsack.nl')

Which generates the file knapsack.nl:

g3 1 1 0        # problem unknown
 4 1 1 0 0      # vars, constraints, objectives, ranges, eqns
 0 0 0 0 0 0    # nonlinear constrs, objs; ccons: lin, nonlin, nd, nzlb
 0 0    # network constraints: nonlinear, linear
 0 0 0  # nonlinear vars in constraints, objectives, both
 0 0 0 1        # linear network variables; functions; arith, flags
 4 0 0 0 0      # discrete variables: binary, integer, nonlinear (b,c,o)
 4 4    # nonzeros in Jacobian, obj. gradient
 0 0    # max name lengths: constraints, variables
 0 0 0 0 0      # common exprs: b,c,o,c1,o1
C0
n0
O0 1
n0
x4
0 True
1 True
2 True
3 True
r
1 14
b
0 0 1
0 0 1
0 0 1
0 0 1
k3
1
2
3
J0 4
0 5
1 4
2 3
3 7
G0 4
0 8
1 6
2 11
3 3

Error Message

SCIP output to scip knapsack.nl -AMPL:

SCIP version 8.0.3 [precision: 8 byte] [memory: block] [mode: optimized] [LP solver: Soplex 6.0.3] [GitHash: 62fab8a2e3]
Copyright (C) 2002-2022 Konrad-Zuse-Zentrum fuer Informationstechnik Berlin (ZIB)

External libraries:
  Soplex 6.0.3         Linear Programming Solver developed at Zuse Institute Berlin (soplex.zib.de) [GitHash: f900e3d0]
  CppAD 20180000.0     Algorithmic Differentiation of C++ algorithms developed by B. Bell (github.com/coin-or/CppAD)
  ZLIB 1.2.13          General purpose compression library by J. Gailly and M. Adler (zlib.net)
  AMPL/MP 4e2d45c4     AMPL .nl file reader library (github.com/ampl/mp)
  PaPILO 2.1.2         parallel presolve for integer and linear optimization (github.com/scipopt/papilo) [GitHash: 2fe2543]
  bliss 0.77           Computing Graph Automorphism Groups by T. Junttila and P. Kaski (www.tcs.hut.fi/Software/bliss/)
  Ipopt 3.14.12        Interior Point Optimizer developed by A. Waechter et.al. (github.com/coin-or/Ipopt)

user parameter file <scip.set> not found - using default parameters
read problem <knapsack.nl>
============

[reader_nl.cpp:1355] ERROR: knapsack.nl:16:3: expected double
[scipshell.c:216] ERROR: Error <-2> in function call
[scipshell.c:272] ERROR: Error <-2> in function call
[scipshell.c:559] ERROR: Error <-2> in function call
SCIP Error (-2): read error

Information on your system

Pyomo version: 6.5.0
Python version: 3.10.10
Operating system: Windows 10
How Pyomo was installed (PyPI, conda, source): conda
Solver (if applicable): SCIP

Additional information

pyomo/repn/plugins/nl_writer.pynl_write.py uses calls to repr() extensively, mostly as f-strings (f'var {value!r}'). Since pyomo stores the values for boolean variables as numpy.bool_ type, the use of repr() alone is not adequate to write their values as SCIP-readable .nl files.

The above case can be fixed by replacing the code block starting at this line:

         _init_lines = [
             f'{var_idx} {info[0].value!r}{col_comments[var_idx]}\n'
             for var_idx, info in enumerate(variables)
             if info[0].value is not None
         ]

by this block:

         _init_lines = []
         for var_idx, info in enumerate(variables):
             if info[0].value is not None:
                 nonbool_value = info[0].value
                 if isinstance(nonbool_value, np.bool_):
                     nonbool_value = int(info[0].value)
                 _init_lines.append(f'{var_idx} {nonbool_value!r}{col_comments[var_idx]}\n')

This is probably not the way to fix this bug. Maybe Boolean variables need to be cast to Binary [0, 1] before calling the writer. Otherwise, there are likely other place in the writer code that would need changing.