Removing hard dependencies

pyomo/contrib/piecewise/tests/test_nonlinear_to_pwl.py

@@ -116,27 +116,28 @@ def test_log_constraint_random_grid(self):
         self.check_pw_linear_log_x(m, pwlf, x1, x2, x3)
 
 
-    def test_log_constraint_lmt_uniform_sample(self):
-        m = self.make_model()
+    # def test_log_constraint_lmt_uniform_sample(self):
+    #     m = self.make_model()
 
-        n_to_pwl = TransformationFactory('contrib.piecewise.nonlinear_to_pwl')
-        n_to_pwl.apply_to(
-            m,
-            num_points=3,
-            domain_partitioning_method=DomainPartitioningMethod.LINEAR_MODEL_TREE_UNIFORM,
-        )
-
-        # cons is transformed
-        self.assertFalse(m.cons.active)
-
-        pwlf = list(m.component_data_objects(PiecewiseLinearFunction,
-                                             descend_into=True))
-        self.assertEqual(len(pwlf), 1)
-        pwlf = pwlf[0]
-
-        set_trace()
-
-        x1 = 4.370861069626263
-        x2 = 7.587945476302646
-        x3 = 9.556428757689245
-        self.check_pw_linear_log_x(m, pwlf, x1, x2, x3)
+    #     n_to_pwl = TransformationFactory('contrib.piecewise.nonlinear_to_pwl')
+    #     n_to_pwl.apply_to(
+    #         m,
+    #         num_points=3,
+    #         domain_partitioning_method=DomainPartitioningMethod.LINEAR_MODEL_TREE_UNIFORM,
+    #     )
+
+    #     # cons is transformed
+    #     self.assertFalse(m.cons.active)
+
+    #     pwlf = list(m.component_data_objects(PiecewiseLinearFunction,
+    #                                          descend_into=True))
+    #     self.assertEqual(len(pwlf), 1)
+    #     pwlf = pwlf[0]
+
+    #     set_trace()
+
+    #     # TODO
+    #     x1 = 4.370861069626263
+    #     x2 = 7.587945476302646
+    #     x3 = 9.556428757689245
+    #     self.check_pw_linear_log_x(m, pwlf, x1, x2, x3)

pyomo/contrib/piecewise/transform/nonlinear_to_pwl.py

@@ -12,10 +12,7 @@
 import enum
 import itertools
 
-from lineartree import LinearTreeRegressor
-import lineartree
 import logging
-import numpy as np
 
 from pyomo.environ import (
     TransformationFactory,
@@ -41,6 +38,8 @@
 from pyomo.common.autoslots import AutoSlots
 from pyomo.common.collections import ComponentMap, ComponentSet
 from pyomo.common.config import ConfigDict, ConfigValue, PositiveInt, InEnum
+from pyomo.common.dependencies import attempt_import
+from pyomo.common.dependencies import numpy as np
 from pyomo.common.modeling import unique_component_name
 from pyomo.core.expr.numeric_expr import SumExpression
 from pyomo.core.expr import identify_variables
@@ -54,13 +53,12 @@
 from pyomo.network import Port
 from pyomo.repn.quadratic import QuadraticRepnVisitor
 
-from sklearn.linear_model import LinearRegression
-import random
-from sklearn.metrics import mean_squared_error
-from sklearn.model_selection import train_test_split
+lineartree, lineartree_available = attempt_import('lineartree')
+sklearn_lm, sklearn_available = attempt_import('sklearn.linear_model')
 
 logger = logging.getLogger(__name__)
 
+
 class DomainPartitioningMethod(enum.IntEnum):
     RANDOM_GRID = 1
     UNIFORM_GRID = 2
@@ -117,22 +115,15 @@ def get_points_lmt(points, bounds, func, seed):
     y_list = []
     for point in points:
         y_list.append(func(*point))
-    regr = LinearTreeRegressor(
-        LinearRegression(),
+    # ESJ: Do we really need the sklearn dependency to get LinearRegression??
+    regr = lineartree.LinearTreeRegressor(
+        sklearn_lm.LinearRegression(),
         criterion='mse',
         max_bins=120,
         min_samples_leaf=4,
         max_depth=5,
     )
-
-    # Using train_test_split is silly. TODO: remove this and just sample my own
-    # extra points if I want to estimate the error.
-    X_train, X_test, y_train, y_test = train_test_split(
-        x_list, y_list, test_size=0.2, random_state=seed
-    )
-    regr.fit(X_train, y_train)
-    y_pred = regr.predict(X_test)
-    error = mean_squared_error(y_test, y_pred)
+    regr.fit(x_list, y_list)
 
     leaves, splits, ths = parse_linear_tree_regressor(regr, bounds)
 
@@ -572,6 +563,7 @@ def _get_bounds_list(self, var_list, parent_component):
         bounds = []
         for v in var_list:
             if None in v.bounds:
+                # ESJ TODO: Con is undefined--this is a bug!
                 raise ValueError(
                     "Cannot automatically approximate constraints with unbounded "
                     "variables. Var '%s' appearining in component '%s' is missing "