Clean mixed integer sampling & seeding (#506)

* fix sampling (notebook and doc remain to do)

* fix ego seeding

* add _to_seed

* initialize seed in design space

smt/applications/ego.py

@@ -14,7 +14,10 @@
 
 from smt.surrogate_models import KPLS, KRG, KPLSK, MGP, GEKPLS
 from smt.applications.application import SurrogateBasedApplication
-from smt.applications.mixed_integer import MixedIntegerContext
+from smt.applications.mixed_integer import (
+    MixedIntegerContext,
+    MixedIntegerSamplingMethod,
+)
 from smt.utils.design_space import (
     BaseDesignSpace,
     DesignSpace,
@@ -271,10 +274,14 @@ def _setup_optimizer(self, fun):
 
         else:
             self.mixint = None
-            self._sampling = lambda n: self.design_space.sample_valid_x(
-                n,
+            sampling = MixedIntegerSamplingMethod(
+                LHS,
+                self.design_space,
+                criterion="ese",
                 random_state=self.options["random_state"],
-            )[0]
+            )
+            self._sampling = lambda n: sampling(n)
+
             self.categorical_kernel = None
 
         # Build DOE

smt/applications/mixed_integer.py

@@ -43,6 +43,10 @@ def __init__(self, sampling_method_class, design_space, **kwargs):
             category=DeprecationWarning,
         )
         self._design_space = design_space
+        if "random_state" in kwargs:
+            self._design_space.random_state = kwargs["random_state"]
+        elif self._design_space.random_state is None:
+            self._design_space.random_state = 42
         self._unfolded_xlimits = design_space.get_unfolded_num_bounds()
         self._output_in_folded_space = kwargs.get("output_in_folded_space", True)
         kwargs.pop("output_in_folded_space", None)
@@ -51,16 +55,20 @@ def __init__(self, sampling_method_class, design_space, **kwargs):
         )
         super().__init__(xlimits=self._unfolded_xlimits)
 
-    def _compute(self, nt):
-        doe = self._sampling_method(nt)
+    def _compute(self, nt, return_is_acting=False):
 
-        x_doe, _ = self._design_space.correct_get_acting(doe)
-        if self._output_in_folded_space:
-            x_doe, _ = self._design_space.fold_x(x_doe)
-        return x_doe
+        x_doe, is_acting = self._design_space.sample_valid_x(
+            nt,
+            unfolded=not self._output_in_folded_space,
+            random_state=self._design_space.random_state,
+        )
+        if return_is_acting:
+            return x_doe, is_acting
+        else:
+            return x_doe
 
-    def __call__(self, nt):
-        return self._compute(nt)
+    def __call__(self, nt, return_is_acting=False):
+        return self._compute(nt, return_is_acting)
 
     def expand_lhs(self, x, nt, method="basic"):
         doe = self._sampling_method(nt)

smt/applications/tests/test_ego.py

@@ -121,7 +121,7 @@ def test_rosenbrock_2D(self):
         xlimits = fun.xlimits
         criterion = "LCB"  #'EI' or 'SBO' or 'LCB'
         random_state = 42
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
         xdoe = FullFactorial(xlimits=xlimits)(10)
         ego = EGO(
             n_start=30,
@@ -164,7 +164,7 @@ def test_rosenbrock_2D_parallel(self):
         xlimits = fun.xlimits
         criterion = "LCB"  #'EI' or 'SBO' or 'LCB'
         random_state = 42
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
 
         xdoe = FullFactorial(xlimits=xlimits)(10)
         qEI = "KB"
@@ -250,7 +250,7 @@ def test_branin_2D_mixed_parallel(self):
                 IntegerVariable(*xlimits[0]),
                 FloatVariable(*xlimits[1]),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
         sm = KRG(design_space=design_space, print_global=False, n_start=25)
         mixint = MixedIntegerContext(design_space)
@@ -289,7 +289,7 @@ def test_branin_2D_mixed(self):
                 IntegerVariable(*xlimits[0]),
                 FloatVariable(*xlimits[1]),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
         criterion = "EI"  #'EI' or 'SBO' or 'LCB'
 
@@ -326,7 +326,7 @@ def test_branin_2D_mixed_tunnel(self):
                 IntegerVariable(*xlimits[0]),
                 FloatVariable(*xlimits[1]),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
         criterion = "EI"  #'EI' or 'SBO' or 'LCB'
 
@@ -381,17 +381,19 @@ def function_test_mixed_integer(X):
     def test_ego_mixed_integer(self):
         n_iter = 15
         n_doe = 5
-        random_state = 42
         design_space = DesignSpace(
             [
                 FloatVariable(-5, 5),
                 CategoricalVariable(["blue", "red", "green"]),
                 CategoricalVariable(["large", "small"]),
                 OrdinalVariable([0, 2, 3]),
             ],
-            seed=random_state,
+            random_state=42,
         )
-        xdoe, _ = design_space.sample_valid_x(n_doe)
+        samp = MixedIntegerSamplingMethod(
+            LHS, design_space, criterion="ese", random_state=design_space.random_state
+        )
+        xdoe = samp(n_doe)
 
         criterion = "EI"  #'EI' or 'SBO' or 'LCB'
         ego = EGO(
@@ -400,7 +402,7 @@ def test_ego_mixed_integer(self):
             xdoe=xdoe,
             surrogate=KRG(design_space=design_space, print_global=False),
             enable_tunneling=False,
-            random_state=random_state,
+            random_state=design_space.random_state,
         )
         _, y_opt, _, _, _ = ego.optimize(fun=TestEGO.function_test_mixed_integer)
 
@@ -418,9 +420,12 @@ def test_ego_mixed_integer_gower_distance(self):
                 CategoricalVariable(["large", "small"]),
                 IntegerVariable(0, 2),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
-        xdoe, _ = design_space.sample_valid_x(n_doe)
+        samp = MixedIntegerSamplingMethod(
+            LHS, design_space, criterion="ese", random_state=design_space.random_state
+        )
+        xdoe = samp(n_doe)
 
         criterion = "EI"  #'EI' or 'SBO' or 'LCB'
         ego = EGO(
@@ -435,7 +440,7 @@ def test_ego_mixed_integer_gower_distance(self):
                 print_global=False,
             ),
             enable_tunneling=False,
-            random_state=random_state,
+            random_state=design_space.random_state,
         )
         _, y_opt, _, _, _ = ego.optimize(fun=TestEGO.function_test_mixed_integer)
 
@@ -490,15 +495,15 @@ def f_hv(X):
                 IntegerVariable(0, 5),  # x6
                 IntegerVariable(0, 5),  # x7
             ],
-            seed=random_state,
+            random_state=random_state,
         )
 
         # x6 is active when x0 >= 2
         design_space.declare_decreed_var(decreed_var=6, meta_var=0, meta_value=[2, 3])
         # x7 is active when x0 >= 3
         design_space.declare_decreed_var(decreed_var=7, meta_var=0, meta_value=3)
 
-        n_doe = 4
+        n_doe = 5
 
         neutral_var_ds = DesignSpace(design_space.design_variables[1:])
         sampling = MixedIntegerSamplingMethod(
@@ -534,7 +539,7 @@ def f_hv(X):
         Xt = np.concatenate((xdoe1, xdoe2, xdoe3), axis=0)
         # Yt = np.concatenate((ydoe1, ydoe2, ydoe3), axis=0)
 
-        n_iter = 6
+        n_iter = 10
         criterion = "EI"
 
         ego = EGO(
@@ -557,7 +562,7 @@ def f_hv(X):
         self.assertAlmostEqual(
             f_hv(np.atleast_2d([2, -5, -5, 5, 0, 0, 0, 5])),
             float(y_opt),
-            delta=15,
+            delta=18,
         )
 
     @unittest.skipIf(int(os.getenv("RUN_SLOW", 0)) < 1, "too slow")
@@ -681,7 +686,7 @@ def f_hv(X):
                 IntegerVariable(0, 2),
                 IntegerVariable(0, 2),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
 
         # x4 is acting if meta == 1, 3
@@ -694,8 +699,10 @@ def f_hv(X):
         ds.declare_decreed_var(decreed_var=8, meta_var=0, meta_value=[0, 1])
 
         n_doe = 25
-        ds.seed = random_state
-        Xt, x_is_active = ds.sample_valid_x(n_doe)
+        samp = MixedIntegerSamplingMethod(
+            LHS, ds, criterion="ese", random_state=ds.random_state
+        )
+        Xt, x_is_active = samp(n_doe, return_is_acting=True)
 
         n_iter = 10
         criterion = "EI"
@@ -735,7 +742,7 @@ def test_ego_mixed_integer_homo_gaussian(self):
                 CategoricalVariable(["large", "small"]),
                 IntegerVariable(0, 2),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
         n_doe = 5
         sampling = MixedIntegerSamplingMethod(
@@ -775,7 +782,7 @@ def test_ego_mixed_integer_homo_gaussian_pls(self):
                 CategoricalVariable(["large", "small"]),
                 IntegerVariable(0, 2),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
         sampling = MixedIntegerSamplingMethod(
             LHS,
@@ -812,7 +819,7 @@ def test_ydoe_option(self):
         xlimits = fun.xlimits
         criterion = "LCB"  #'EI' or 'SBO' or 'LCB'
         random_state = 42
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
         xdoe = FullFactorial(xlimits=xlimits)(10)
         ydoe = fun(xdoe)
         ego = EGO(
@@ -831,7 +838,7 @@ def test_find_best_point(self):
         fun = TestEGO.function_test_1d
         xlimits = np.array([[0.0, 25.0]])
         random_state = 42
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
         xdoe = FullFactorial(xlimits=xlimits)(3)
         ydoe = fun(xdoe)
         ego = EGO(
@@ -867,7 +874,7 @@ def _evaluate(self, x, kx):
 
         fun = TensorProductIndirect(ndim=2, func=func)
         random_state = 42
-        design_space = DesignSpace(fun.xlimits, seed=42)
+        design_space = DesignSpace(fun.xlimits, random_state=42)
 
         # Construction of the DOE
         sampling = LHS(xlimits=fun.xlimits, criterion="m", random_state=random_state)
@@ -899,7 +906,7 @@ def _evaluate(self, x, kx):
 
         return ego, fun
 
-    def test_ego_seeding(self):
+    def test_ego_random_stateing(self):
         def f_obj(X):
             """
             s01 objective
@@ -976,7 +983,7 @@ def f_obj(X):
         )
 
         # To define the initial DOE
-        random_state = 42  # seed value for the sampling
+        random_state = 42  # random_state value for the sampling
         n_doe = 5  # initial doe size
         sampling = MixedIntegerSamplingMethod(
             LHS, design_space, criterion="ese", random_state=random_state
@@ -1005,15 +1012,11 @@ def f_obj(X):
         )
         x_opt, y_opt, dnk, x_data, y_data = ego.optimize(fun=f_obj)
         if ds.HAS_CONFIG_SPACE:  # results differs wrt config_space impl
-            if platform.startswith("linux"):  # results differs wrt platform
-                self.assertAlmostEqual(np.sum(y_data), 1.0355815090110578, delta=1e-12)
-                self.assertAlmostEqual(np.sum(x_data), 38.56885202767958, delta=1e-12)
-            else:
-                self.assertAlmostEqual(np.sum(y_data), 0.9606415626557894, delta=1e-12)
-                self.assertAlmostEqual(np.sum(x_data), 38.23494224077761, delta=1e-12)
+            self.assertAlmostEqual(np.sum(y_data), 6.846225752638086, delta=1e-9)
+            self.assertAlmostEqual(np.sum(x_data), 33.81192549170815, delta=1e-9)
         else:
-            self.assertAlmostEqual(np.sum(y_data), 1.8911720770059735, delta=1e-12)
-            self.assertAlmostEqual(np.sum(x_data), 47.56885202767958, delta=1e-12)
+            self.assertAlmostEqual(np.sum(y_data), 1.8911720670620835, delta=1e-9)
+            self.assertAlmostEqual(np.sum(x_data), 47.56885202767958, delta=1e-9)
 
     def test_ego_gek(self):
         ego, fun = self.initialize_ego_gek()
@@ -1036,7 +1039,7 @@ def test_qei_criterion_default(self):
         fun = TestEGO.function_test_1d
         xlimits = np.array([[0.0, 25.0]])
         random_state = 42
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
         xdoe = FullFactorial(xlimits=xlimits)(3)
         ydoe = fun(xdoe)
         ego = EGO(
@@ -1091,7 +1094,7 @@ def function_test_1d(x):
         xlimits = np.array([[0.0, 25.0]])
 
         random_state = 42  # for reproducibility
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
         xdoe = np.atleast_2d([0, 7, 25]).T
         n_doe = xdoe.size
 
@@ -1210,7 +1213,7 @@ def function_test_mixed_integer(X):
                 CategoricalVariable(["square", "circle"]),
                 IntegerVariable(0, 2),
             ],
-            seed=random_state,
+            random_state=random_state,
         )
 
         criterion = "EI"  #'EI' or 'SBO' or 'LCB'
@@ -1281,7 +1284,7 @@ def function_test_1d(x):
         xlimits = np.array([[0.0, 25.0]])
 
         random_state = 42
-        design_space = DesignSpace(xlimits, seed=random_state)
+        design_space = DesignSpace(xlimits, random_state=random_state)
         xdoe = np.atleast_2d([0, 7, 25]).T
         n_doe = xdoe.size