Quadratic term not identified by generate_standard_repn

[michaelbynum]

>>> import pyomo.environ as pe                                                                                                                                                                                               
>>> from pyomo.repn.standard_repn import generate_standard_repn
>>> m = pe.ConcreteModel()
>>> m.x = pe.Var()
>>> m.y = pe.Var()
>>> print(generate_standard_repn((m.x + m.y)**2, quadratic=True))

constant:       0
linear vars:    []
linear var ids: []
linear coef:    []
quadratic vars:    []
quadratic var ids: []
quadratic coef:    []
nonlinear expr:
(x + y)**2
nonlinear vars: ['x', 'y']

[carldlaird]

@michaelbynum good catch. We should also consider adding an explicit QuadraticExpression object like we have for LinearExpression.

[johwiebe]

@michaelbynum I also just came across this issue. I noticed that there is a test which asks for this behaviour. Is this by design for some reason?

pyomo/pyomo/repn/tests/test_standard.py

Lines 3121 to 3157 in 7920516

	e = (m.a+m.b)**2
	rep = generate_standard_repn(e, compute_values=False, quadratic=False)
	#
	self.assertFalse( rep.is_fixed() )
	self.assertEqual( rep.polynomial_degree(), None )
	self.assertFalse( rep.is_constant() )
	self.assertFalse( rep.is_linear() )
	self.assertFalse( rep.is_quadratic() )
	self.assertTrue( rep.is_nonlinear() )
	#
	self.assertTrue(len(rep.linear_vars) == 0)
	self.assertTrue(len(rep.linear_coefs) == 0)
	self.assertTrue(len(rep.quadratic_vars) == 0)
	self.assertTrue(len(rep.quadratic_coefs) == 0)
	self.assertFalse(rep.nonlinear_expr is None)
	self.assertTrue(len(rep.nonlinear_vars) == 2)
	baseline = { }
	self.assertEqual(baseline, repn_to_dict(rep))
	rep = generate_standard_repn(e, compute_values=False, quadratic=True)
	#
	self.assertFalse( rep.is_fixed() )
	self.assertEqual( rep.polynomial_degree(), None )
	self.assertFalse( rep.is_constant() )
	self.assertFalse( rep.is_linear() )
	self.assertFalse( rep.is_quadratic() )
	self.assertTrue( rep.is_nonlinear() )
	#
	self.assertTrue(len(rep.linear_vars) == 0)
	self.assertTrue(len(rep.linear_coefs) == 0)
	self.assertTrue(len(rep.quadratic_vars) == 0)
	self.assertTrue(len(rep.quadratic_coefs) == 0)
	self.assertFalse(rep.nonlinear_expr is None)
	self.assertTrue(len(rep.nonlinear_vars) == 2)
	baseline = { }
	self.assertEqual(baseline, repn_to_dict(rep))

[michaelbynum]

@johwiebe The test you mentioned does not actually expect this behavior. The key is the quadratic=False keyword argument to the generate_standard_repn function. When quadratic=False, quadratic terms are not identified (only linear).

[johwiebe]

@michaelbynum I think that's just the first part of the test. On line 3141 the same expression is tested with quadratic=True:

pyomo/pyomo/repn/tests/test_standard.py

Lines 3141 to 3157 in 7920516

	rep = generate_standard_repn(e, compute_values=False, quadratic=True)
	#
	self.assertFalse( rep.is_fixed() )
	self.assertEqual( rep.polynomial_degree(), None )
	self.assertFalse( rep.is_constant() )
	self.assertFalse( rep.is_linear() )
	self.assertFalse( rep.is_quadratic() )
	self.assertTrue( rep.is_nonlinear() )
	#
	self.assertTrue(len(rep.linear_vars) == 0)
	self.assertTrue(len(rep.linear_coefs) == 0)
	self.assertTrue(len(rep.quadratic_vars) == 0)
	self.assertTrue(len(rep.quadratic_coefs) == 0)
	self.assertFalse(rep.nonlinear_expr is None)
	self.assertTrue(len(rep.nonlinear_vars) == 2)
	baseline = { }
	self.assertEqual(baseline, repn_to_dict(rep))

The expected outcome with and without quadratic=False seems to be the same.

[michaelbynum]

@johwiebe Oh, I apologize. I did not scroll far enough. Good catch. I don't think this behavior is by design.