Fix noisy categorical PLS bug (#454)

SMTorg · Jul 10, 2023 · 899b7d0 · 899b7d0
1 parent e4e35ce
commit 899b7d0
Show file tree

Hide file tree

Showing 2 changed files with 42 additions and 2 deletions.
diff --git a/smt/applications/tests/test_mixed_integer.py b/smt/applications/tests/test_mixed_integer.py
@@ -1017,6 +1017,46 @@ def test_mixed_CR_2D(self):
 
         self.assertEqual(np.shape(y), (105, 1))
 
+    def test_mixed_CR_PLS_noisy_2D(self):
+        xt = np.array([[0, 5], [2, -1], [4, 0.5]])
+        yt = np.array([[0.0], [1.0], [1.5]])
+        design_space = DesignSpace(
+            [
+                CategoricalVariable(["0.0", "1.0", " 2.0", "3.0", "4.0"]),
+                FloatVariable(-5, 5),
+            ]
+        )
+        # Surrogate
+        sm = MixedIntegerKrigingModel(
+            surrogate=KPLS(
+                n_comp=1,
+                eval_noise=True,
+                design_space=design_space,
+                theta0=[1e-2],
+                categorical_kernel=MixIntKernelType.CONT_RELAX,
+                corr="abs_exp",
+            ),
+        )
+        sm.set_training_values(xt, yt)
+        sm.train()
+
+        # DOE for validation
+        x = np.linspace(0, 4, 5)
+        x2 = np.linspace(-5, 5, 21)
+        x1 = []
+        for element in itertools.product(x, x2):
+            x1.append(np.array(element))
+        x_pred = np.array(x1)
+
+        y = sm.predict_values(x_pred)
+        yvar = sm.predict_variances(x_pred)
+
+        # prediction are correct on known points
+        self.assertTrue(np.abs(np.sum(np.array([y[20], y[50], y[95]]) - yt)) < 1e-6)
+        self.assertTrue(np.abs(np.sum(np.array([yvar[20], yvar[50], yvar[95]]))) < 1e-6)
+
+        self.assertEqual(np.shape(y), (105, 1))
+
     def test_mixed_homo_gaussian_2D(self):
         xt = np.array([[0, 5], [2, -1], [4, 0.5]])
         yt = np.array([[0.0], [1.0], [1.5]])

diff --git a/smt/surrogate_models/krg_based.py b/smt/surrogate_models/krg_based.py
@@ -698,8 +698,8 @@ def _reduced_likelihood_function(self, theta):
         if self.options["use_het_noise"]:
             noise = self.optimal_noise
         if self.options["eval_noise"] and not self.options["use_het_noise"]:
-            theta = tmp_var[0 : self.D.shape[1]]
-            noise = tmp_var[self.D.shape[1] :]
+            theta = tmp_var[0:-1]
+            noise = tmp_var[-1]
         if not (self.is_continuous):
             dx = self.D
             r = self._matrix_data_corr(