Fix kriging-based surrogates predict_variances when noise0 option…

… is given as a `np.ndarray` (#710)

* Add test to show predict_variances bug with ndarray as noise0

* Fix the test for null noise

* Fix doc typo

* Format with ruff 0.9.2

* Bump ruff version 0.9.2 in pre-commit conf

.pre-commit-config.yaml

@@ -1,7 +1,7 @@
 repos:
 - repo: https://github.com/astral-sh/ruff-pre-commit
   # Ruff version.
-  rev: v0.8.6
+  rev: v0.9.2
   hooks:
     # Run the linter.
     - id: ruff

doc/_src_docs/surrogate_models/gpr.rstx

@@ -1,5 +1,5 @@
-Gausian process regression
-==========================
+Gaussian process regression
+===========================
 
 SMT implements several surrogate models related to Gaussian process regression:
 

smt/applications/ego.py

@@ -181,7 +181,7 @@ def optimize(self, fun):
             end_time = time.time()
             iteration_time = end_time - start_time
             if self.options["verbose"]:
-                print(f"iteration {k+1} took {iteration_time:.2f} seconds")
+                print(f"iteration {k + 1} took {iteration_time:.2f} seconds")
         # Find the optimal point
         ind_best = np.argmin(y_data if y_data.ndim == 1 else y_data[:, 0])
         x_opt = x_data[ind_best]

smt/applications/podi.py

@@ -376,7 +376,7 @@ def interp_POD_basis(
                     f"WARNING: Sample of size {n_real} at {n_new} points of a {self.n_DoF}*{self.n_modes} matrix"
                 )
                 print(
-                    f"Be sure to get sufficient memory for {n_real*n_new*self.n_DoF*self.n_modes} floats"
+                    f"Be sure to get sufficient memory for {n_real * n_new * self.n_DoF * self.n_modes} floats"
                 )
             # compute the interpolation of the coefficients and the associated variance
             coeff, var = self.pred_coeff(mu, compute_var=True)

smt/surrogate_models/krg_based.py

@@ -1859,7 +1859,7 @@ def _predict_variances(
             np.dot(self.optimal_par["Ft"].T, rt)
             - self._regression_types[self.options["poly"]](X_cont).T,
         )
-        is_noisy = self.options["noise0"] != [0.0] or self.options["eval_noise"]
+        is_noisy = np.max(self.options["noise0"]) > 0.0 or self.options["eval_noise"]
         if is_noisy and is_ri:
             A = self.optimal_par["sigma2_ri"]
         else:

smt/surrogate_models/tests/test_krg_noise.py

@@ -54,6 +54,11 @@ def test_predict_variance(self):
             noise_bounds=[1e-2, 1000.0],
             print_global=False,
         )  # noisy Kriging model with estimated variance
+        sm_het_noise_estim = KRG(
+            noise0=np.array([1e-2, 2e-2, 3e-2, 4e-2, 5e-2]),  # test with numpy array
+            use_het_noise=True,
+            print_global=False,
+        )  # heteroscedastic noisy Kriging model
 
         # training the models
         sm_noise_free.set_training_values(xt, yt)
@@ -65,6 +70,9 @@ def test_predict_variance(self):
         sm_noise_estim.set_training_values(xt, yt)
         sm_noise_estim.train()
 
+        sm_het_noise_estim.set_training_values(xt, yt)
+        sm_het_noise_estim.train()
+
         # predictions at training points
         x = xt
 
@@ -79,6 +87,10 @@ def test_predict_variance(self):
         self.assert_error(np.linalg.norm(var_noise_fixed), 0.04768, 1e-5)
         var_noise_estim = sm_noise_estim.predict_variances(x)  # predictive variance
         self.assert_error(np.linalg.norm(var_noise_estim), 0.01135, 1e-3)
+        var_het_noise_estim = sm_het_noise_estim.predict_variances(
+            x
+        )  # predictive variance
+        self.assert_error(np.linalg.norm(var_het_noise_estim), 0.02, 1e-3)
 
     def test_null_predict_variance_ri_at_data_points(self):
         # defining the training data

tutorial/Misc/SMT_SGP_analytic.ipynb

@@ -1139,7 +1139,7 @@
     "print(\n",
     "    \"Considering float64, memory required: ~ %.2f Go\" % (8 * xt.shape[0] ** 2 / 1024**3)\n",
     ")\n",
-    "print(f\"Inference cost: ~ {xt.shape[0]**3:.2e} operations\")"
+    "print(f\"Inference cost: ~ {xt.shape[0] ** 3:.2e} operations\")"
    ]
   },
   {
@@ -1176,7 +1176,7 @@
     "print(\n",
     "    \"Considering float64, memory required: ~ %.2f Mo\" % (8 * 50 * xt.shape[0] / 1024**2)\n",
     ")\n",
-    "print(f\"Inference cost: ~ {xt.shape[0]*50**2:.2e} operations\")"
+    "print(f\"Inference cost: ~ {xt.shape[0] * 50**2:.2e} operations\")"
    ]
   },
   {

tutorial/PODI/SMT_PODI_Airfoil.ipynb

@@ -462,7 +462,7 @@
     "        contour_gt = ax.tricontourf(\n",
     "            triang, P_test_list[idx].flatten(), levels=100, cmap=\"viridis\"\n",
     "        )\n",
-    "        ax.set_title(f\"Ground Truth Sample {idx+1}\")\n",
+    "        ax.set_title(f\"Ground Truth Sample {idx + 1}\")\n",
     "        ax.set_xlabel(\"X Coordinate\")\n",
     "        ax.set_ylabel(\"Y Coordinate\")\n",
     "        fig.colorbar(contour_gt, ax=ax, orientation=\"vertical\")\n",
@@ -472,7 +472,7 @@
     "        contour_pred = ax.tricontourf(\n",
     "            triang, P_pred_list[idx].flatten(), levels=100, cmap=\"viridis\"\n",
     "        )\n",
-    "        ax.set_title(f\"Predicted PODI Pressure Sample {idx+1}\")\n",
+    "        ax.set_title(f\"Predicted PODI Pressure Sample {idx + 1}\")\n",
     "        ax.set_xlabel(\"X Coordinate\")\n",
     "        fig.colorbar(contour_pred, ax=ax, orientation=\"vertical\")\n",
     "\n",

tutorial/SBO/SMT_EGO_noisyGP.ipynb

@@ -817,10 +817,10 @@
    ],
    "source": [
     "print(\n",
-    "    f\"with re-interpolation we get a variance at xdoe < 0.1 {t_noisy.predict_variances(xdoe, is_ri=True)<1e-1}\"\n",
+    "    f\"with re-interpolation we get a variance at xdoe < 0.1 {t_noisy.predict_variances(xdoe, is_ri=True) < 1e-1}\"\n",
     ")\n",
     "print(\n",
-    "    f\"without re-interpolation we get a variance at xdoe > 1 {t_noisy.predict_variances(xdoe, is_ri=False)>1}\"\n",
+    "    f\"without re-interpolation we get a variance at xdoe > 1 {t_noisy.predict_variances(xdoe, is_ri=False) > 1}\"\n",
     ")"
    ]
   },