Merge pull request #3036 from sadavis1/logarithmic_pwlinear

Piecewise Linear transformations: Logarithmic and nested inner GDP representation

Author: blnicho

pyomo/contrib/piecewise/__init__.py

@@ -33,3 +33,9 @@
 from pyomo.contrib.piecewise.transform.convex_combination import (
     ConvexCombinationTransformation,
 )
+from pyomo.contrib.piecewise.transform.nested_inner_repn import (
+    NestedInnerRepresentationGDPTransformation,
+)
+from pyomo.contrib.piecewise.transform.disaggregated_logarithmic import (
+    DisaggregatedLogarithmicMIPTransformation,
+)

pyomo/contrib/piecewise/tests/common_inner_repn_tests.py

@@ -0,0 +1,80 @@
+#  ___________________________________________________________________________
+#
+#  Pyomo: Python Optimization Modeling Objects
+#  Copyright (c) 2008-2024
+#  National Technology and Engineering Solutions of Sandia, LLC
+#  Under the terms of Contract DE-NA0003525 with National Technology and
+#  Engineering Solutions of Sandia, LLC, the U.S. Government retains certain
+#  rights in this software.
+#  This software is distributed under the 3-clause BSD License.
+#  ___________________________________________________________________________
+
+from pyomo.core import Var
+from pyomo.core.base import Constraint
+from pyomo.core.expr.compare import assertExpressionsEqual
+
+# This file contains check methods shared between GDP inner representation-based
+# transformations. Currently, those are the inner_representation_gdp and
+# nested_inner_repn_gdp transformations, since each have disjuncts with the
+# same structure.
+
+
+# Check one disjunct from the log model for proper contents
+def check_log_disjunct(test, d, pts, f, substitute_var, x):
+    test.assertEqual(len(d.component_map(Constraint)), 3)
+    # lambdas and indicator_var
+    test.assertEqual(len(d.component_map(Var)), 2)
+    test.assertIsInstance(d.lambdas, Var)
+    test.assertEqual(len(d.lambdas), 2)
+    for lamb in d.lambdas.values():
+        test.assertEqual(lamb.lb, 0)
+        test.assertEqual(lamb.ub, 1)
+    test.assertIsInstance(d.convex_combo, Constraint)
+    assertExpressionsEqual(test, d.convex_combo.expr, d.lambdas[0] + d.lambdas[1] == 1)
+    test.assertIsInstance(d.set_substitute, Constraint)
+    assertExpressionsEqual(
+        test, d.set_substitute.expr, substitute_var == f(x), places=7
+    )
+    test.assertIsInstance(d.linear_combo, Constraint)
+    test.assertEqual(len(d.linear_combo), 1)
+    assertExpressionsEqual(
+        test, d.linear_combo[0].expr, x == pts[0] * d.lambdas[0] + pts[1] * d.lambdas[1]
+    )
+
+
+# Check one disjunct from the paraboloid model for proper contents.
+def check_paraboloid_disjunct(test, d, pts, f, substitute_var, x1, x2):
+    test.assertEqual(len(d.component_map(Constraint)), 3)
+    # lambdas and indicator_var
+    test.assertEqual(len(d.component_map(Var)), 2)
+    test.assertIsInstance(d.lambdas, Var)
+    test.assertEqual(len(d.lambdas), 3)
+    for lamb in d.lambdas.values():
+        test.assertEqual(lamb.lb, 0)
+        test.assertEqual(lamb.ub, 1)
+    test.assertIsInstance(d.convex_combo, Constraint)
+    assertExpressionsEqual(
+        test, d.convex_combo.expr, d.lambdas[0] + d.lambdas[1] + d.lambdas[2] == 1
+    )
+    test.assertIsInstance(d.set_substitute, Constraint)
+    assertExpressionsEqual(
+        test, d.set_substitute.expr, substitute_var == f(x1, x2), places=7
+    )
+    test.assertIsInstance(d.linear_combo, Constraint)
+    test.assertEqual(len(d.linear_combo), 2)
+    assertExpressionsEqual(
+        test,
+        d.linear_combo[0].expr,
+        x1
+        == pts[0][0] * d.lambdas[0]
+        + pts[1][0] * d.lambdas[1]
+        + pts[2][0] * d.lambdas[2],
+    )
+    assertExpressionsEqual(
+        test,
+        d.linear_combo[1].expr,
+        x2
+        == pts[0][1] * d.lambdas[0]
+        + pts[1][1] * d.lambdas[1]
+        + pts[2][1] * d.lambdas[2],
+    )

pyomo/contrib/piecewise/tests/test_disaggregated_logarithmic.py

@@ -0,0 +1,275 @@
+#  ___________________________________________________________________________
+#
+#  Pyomo: Python Optimization Modeling Objects
+#  Copyright (c) 2008-2024
+#  National Technology and Engineering Solutions of Sandia, LLC
+#  Under the terms of Contract DE-NA0003525 with National Technology and
+#  Engineering Solutions of Sandia, LLC, the U.S. Government retains certain
+#  rights in this software.
+#  This software is distributed under the 3-clause BSD License.
+#  ___________________________________________________________________________
+
+import pyomo.common.unittest as unittest
+from pyomo.contrib.piecewise.tests import models
+import pyomo.contrib.piecewise.tests.common_tests as ct
+from pyomo.core.base import TransformationFactory
+from pyomo.environ import SolverFactory, Var, Constraint
+from pyomo.core.expr.compare import assertExpressionsEqual
+
+
+class TestTransformPiecewiseModelToNestedInnerRepnMIP(unittest.TestCase):
+    def check_pw_log(self, m):
+        z = m.pw_log.get_transformation_var(m.log_expr)
+        self.assertIsInstance(z, Var)
+        # Now we can use those Vars to check on what the transformation created
+        log_block = z.parent_block()
+
+        # We should have three Vars, two of which are indexed, and five
+        # Constraints, three of which are indexed
+
+        self.assertEqual(len(log_block.component_map(Var)), 3)
+        self.assertEqual(len(log_block.component_map(Constraint)), 5)
+
+        # Constants
+        simplex_count = 3
+        log_simplex_count = 2
+        simplex_point_count = 2
+
+        # Substitute var
+        self.assertIsInstance(log_block.substitute_var, Var)
+        self.assertIs(m.obj.expr.expr, log_block.substitute_var)
+        # Binaries
+        self.assertIsInstance(log_block.binaries, Var)
+        self.assertEqual(len(log_block.binaries), log_simplex_count)
+        # Lambdas
+        self.assertIsInstance(log_block.lambdas, Var)
+        self.assertEqual(len(log_block.lambdas), simplex_count * simplex_point_count)
+        for l in log_block.lambdas.values():
+            self.assertEqual(l.lb, 0)
+            self.assertEqual(l.ub, 1)
+
+        # Convex combo constraint
+        self.assertIsInstance(log_block.convex_combo, Constraint)
+        assertExpressionsEqual(
+            self,
+            log_block.convex_combo.expr,
+            log_block.lambdas[0, 0]
+            + log_block.lambdas[0, 1]
+            + log_block.lambdas[1, 0]
+            + log_block.lambdas[1, 1]
+            + log_block.lambdas[2, 0]
+            + log_block.lambdas[2, 1]
+            == 1,
+        )
+
+        # Set substitute constraint
+        self.assertIsInstance(log_block.set_substitute, Constraint)
+        assertExpressionsEqual(
+            self,
+            log_block.set_substitute.expr,
+            log_block.substitute_var
+            == log_block.lambdas[0, 0] * m.f1(1)
+            + log_block.lambdas[1, 0] * m.f2(3)
+            + log_block.lambdas[2, 0] * m.f3(6)
+            + log_block.lambdas[0, 1] * m.f1(3)
+            + log_block.lambdas[1, 1] * m.f2(6)
+            + log_block.lambdas[2, 1] * m.f3(10),
+            places=7,
+        )
+
+        # x constraint
+        self.assertIsInstance(log_block.x_constraint, Constraint)
+        # one-dimensional case, so there is only one x variable here
+        self.assertEqual(len(log_block.x_constraint), 1)
+        assertExpressionsEqual(
+            self,
+            log_block.x_constraint[0].expr,
+            m.x
+            == 1 * log_block.lambdas[0, 0]
+            + 3 * log_block.lambdas[0, 1]
+            + 3 * log_block.lambdas[1, 0]
+            + 6 * log_block.lambdas[1, 1]
+            + 6 * log_block.lambdas[2, 0]
+            + 10 * log_block.lambdas[2, 1],
+        )
+
+        # simplex choice 1 constraint enables lambdas when binaries are on
+        self.assertEqual(len(log_block.simplex_choice_1), log_simplex_count)
+        assertExpressionsEqual(
+            self,
+            log_block.simplex_choice_1[0].expr,
+            log_block.lambdas[2, 0] + log_block.lambdas[2, 1] <= log_block.binaries[0],
+        )
+        assertExpressionsEqual(
+            self,
+            log_block.simplex_choice_1[1].expr,
+            log_block.lambdas[1, 0] + log_block.lambdas[1, 1] <= log_block.binaries[1],
+        )
+        # simplex choice 2 constraint enables lambdas when binaries are off
+        self.assertEqual(len(log_block.simplex_choice_2), log_simplex_count)
+        assertExpressionsEqual(
+            self,
+            log_block.simplex_choice_2[0].expr,
+            log_block.lambdas[0, 0]
+            + log_block.lambdas[0, 1]
+            + log_block.lambdas[1, 0]
+            + log_block.lambdas[1, 1]
+            <= 1 - log_block.binaries[0],
+        )
+        assertExpressionsEqual(
+            self,
+            log_block.simplex_choice_2[1].expr,
+            log_block.lambdas[0, 0]
+            + log_block.lambdas[0, 1]
+            + log_block.lambdas[2, 0]
+            + log_block.lambdas[2, 1]
+            <= 1 - log_block.binaries[1],
+        )
+
+    def check_pw_paraboloid(self, m):
+        # This is a little larger, but at least test that the right numbers of
+        # everything are created
+        z = m.pw_paraboloid.get_transformation_var(m.paraboloid_expr)
+        self.assertIsInstance(z, Var)
+        paraboloid_block = z.parent_block()
+
+        self.assertEqual(len(paraboloid_block.component_map(Var)), 3)
+        self.assertEqual(len(paraboloid_block.component_map(Constraint)), 5)
+
+        # Constants
+        simplex_count = 4
+        log_simplex_count = 2
+        simplex_point_count = 3
+
+        # Substitute var
+        self.assertIsInstance(paraboloid_block.substitute_var, Var)
+        # Binaries
+        self.assertIsInstance(paraboloid_block.binaries, Var)
+        self.assertEqual(len(paraboloid_block.binaries), log_simplex_count)
+        # Lambdas
+        self.assertIsInstance(paraboloid_block.lambdas, Var)
+        self.assertEqual(
+            len(paraboloid_block.lambdas), simplex_count * simplex_point_count
+        )
+        for l in paraboloid_block.lambdas.values():
+            self.assertEqual(l.lb, 0)
+            self.assertEqual(l.ub, 1)
+
+        # Convex combo constraint
+        self.assertIsInstance(paraboloid_block.convex_combo, Constraint)
+        assertExpressionsEqual(
+            self,
+            paraboloid_block.convex_combo.expr,
+            paraboloid_block.lambdas[0, 0]
+            + paraboloid_block.lambdas[0, 1]
+            + paraboloid_block.lambdas[0, 2]
+            + paraboloid_block.lambdas[1, 0]
+            + paraboloid_block.lambdas[1, 1]
+            + paraboloid_block.lambdas[1, 2]
+            + paraboloid_block.lambdas[2, 0]
+            + paraboloid_block.lambdas[2, 1]
+            + paraboloid_block.lambdas[2, 2]
+            + paraboloid_block.lambdas[3, 0]
+            + paraboloid_block.lambdas[3, 1]
+            + paraboloid_block.lambdas[3, 2]
+            == 1,
+        )
+
+        # Set substitute constraint
+        self.assertIsInstance(paraboloid_block.set_substitute, Constraint)
+        assertExpressionsEqual(
+            self,
+            paraboloid_block.set_substitute.expr,
+            paraboloid_block.substitute_var
+            == paraboloid_block.lambdas[0, 0] * m.g1(0, 1)
+            + paraboloid_block.lambdas[1, 0] * m.g1(0, 1)
+            + paraboloid_block.lambdas[2, 0] * m.g2(3, 4)
+            + paraboloid_block.lambdas[3, 0] * m.g2(0, 7)
+            + paraboloid_block.lambdas[0, 1] * m.g1(0, 4)
+            + paraboloid_block.lambdas[1, 1] * m.g1(3, 4)
+            + paraboloid_block.lambdas[2, 1] * m.g2(3, 7)
+            + paraboloid_block.lambdas[3, 1] * m.g2(0, 4)
+            + paraboloid_block.lambdas[0, 2] * m.g1(3, 4)
+            + paraboloid_block.lambdas[1, 2] * m.g1(3, 1)
+            + paraboloid_block.lambdas[2, 2] * m.g2(0, 7)
+            + paraboloid_block.lambdas[3, 2] * m.g2(3, 4),
+            places=7,
+        )
+
+        # x constraint
+        self.assertIsInstance(paraboloid_block.x_constraint, Constraint)
+        # Here we have two x variables
+        self.assertEqual(len(paraboloid_block.x_constraint), 2)
+        assertExpressionsEqual(
+            self,
+            paraboloid_block.x_constraint[0].expr,
+            m.x1
+            == 0 * paraboloid_block.lambdas[0, 0]
+            + 0 * paraboloid_block.lambdas[0, 1]
+            + 3 * paraboloid_block.lambdas[0, 2]
+            + 0 * paraboloid_block.lambdas[1, 0]
+            + 3 * paraboloid_block.lambdas[1, 1]
+            + 3 * paraboloid_block.lambdas[1, 2]
+            + 3 * paraboloid_block.lambdas[2, 0]
+            + 3 * paraboloid_block.lambdas[2, 1]
+            + 0 * paraboloid_block.lambdas[2, 2]
+            + 0 * paraboloid_block.lambdas[3, 0]
+            + 0 * paraboloid_block.lambdas[3, 1]
+            + 3 * paraboloid_block.lambdas[3, 2],
+        )
+        assertExpressionsEqual(
+            self,
+            paraboloid_block.x_constraint[1].expr,
+            m.x2
+            == 1 * paraboloid_block.lambdas[0, 0]
+            + 4 * paraboloid_block.lambdas[0, 1]
+            + 4 * paraboloid_block.lambdas[0, 2]
+            + 1 * paraboloid_block.lambdas[1, 0]
+            + 4 * paraboloid_block.lambdas[1, 1]
+            + 1 * paraboloid_block.lambdas[1, 2]
+            + 4 * paraboloid_block.lambdas[2, 0]
+            + 7 * paraboloid_block.lambdas[2, 1]
+            + 7 * paraboloid_block.lambdas[2, 2]
+            + 7 * paraboloid_block.lambdas[3, 0]
+            + 4 * paraboloid_block.lambdas[3, 1]
+            + 4 * paraboloid_block.lambdas[3, 2],
+        )
+
+        # The choices will get long, so let's just assert we have enough
+        self.assertEqual(len(paraboloid_block.simplex_choice_1), log_simplex_count)
+        self.assertEqual(len(paraboloid_block.simplex_choice_2), log_simplex_count)
+
+    # Test methods using the common_tests.py code.
+    def test_transformation_do_not_descend(self):
+        ct.check_transformation_do_not_descend(
+            self, 'contrib.piecewise.disaggregated_logarithmic'
+        )
+
+    def test_transformation_PiecewiseLinearFunction_targets(self):
+        ct.check_transformation_PiecewiseLinearFunction_targets(
+            self, 'contrib.piecewise.disaggregated_logarithmic'
+        )
+
+    def test_descend_into_expressions(self):
+        ct.check_descend_into_expressions(
+            self, 'contrib.piecewise.disaggregated_logarithmic'
+        )
+
+    def test_descend_into_expressions_constraint_target(self):
+        ct.check_descend_into_expressions_constraint_target(
+            self, 'contrib.piecewise.disaggregated_logarithmic'
+        )
+
+    def test_descend_into_expressions_objective_target(self):
+        ct.check_descend_into_expressions_objective_target(
+            self, 'contrib.piecewise.disaggregated_logarithmic'
+        )
+
+    # Check solution of the log(x) model
+    @unittest.skipUnless(SolverFactory('gurobi').available(), 'Gurobi is not available')
+    @unittest.skipUnless(SolverFactory('gurobi').license_is_valid(), 'No license')
+    def test_solve_log_model(self):
+        m = models.make_log_x_model()
+        TransformationFactory("contrib.piecewise.disaggregated_logarithmic").apply_to(m)
+        SolverFactory("gurobi").solve(m)
+        ct.check_log_x_model_soln(self, m)