Merge branch 'main' of github.com:sequential-parameter-optimization/spotPython

Author: bartzbeielstein

src/spotpython/fun/xai_hyperlight.py

@@ -169,15 +169,15 @@ def fun(self, X: np.ndarray, fun_control: dict = None) -> np.ndarray:
             # Multiply results by the weights. Positive weights mean that the result is to be minimized.
             # Negative weights mean that the result is to be maximized, e.g., accuracy.
             z_val = fun_control["weights"] * df_eval
-            xai_incons = fun_control["xai_weight"] * xai_attr
 
             # Append, since several configurations can be evaluated at once.
             z_res = np.append(z_res, z_val)
-            xai_res = np.append(xai_res, xai_incons)
+            xai_res = np.append(xai_res, xai_attr)
 
-            print("Performance loss: ", z_val)
-            print("XAI inconsistency: ", xai_incons)
+            # Combine z_res and xai_res into a single array
+            combined_res = np.column_stack((z_res, xai_res))
 
-            res = z_res + xai_res
+            # print("combined: ", combined_res)
+            # print("shape: ", combined_res.shape)
 
-        return res, z_res, xai_res
+        return combined_res

src/spotpython/light/trainmodel.py

@@ -673,23 +673,20 @@ def train_model_xai(config: dict, fun_control: dict, timestamp: bool = True) ->
         attribution_ig = attr_ig.attribute(X_val_tensor, baselines=baseline)
         ig_attr_test_sum = attribution_ig.detach().numpy().sum(0)
         ig_attr_test_norm_sum = ig_attr_test_sum / np.linalg.norm(ig_attr_test_sum, ord=1)
-        print("Integrated Gradients attribution sum:", ig_attr_test_norm_sum)
         attributions_dict["IntegratedGradients"] = ig_attr_test_norm_sum
 
     if "KernelShap" in fun_control["xai_methods"]:
         attr_ks = KernelShap(model)
         attribution_ks = attr_ks.attribute(X_val_tensor, baselines=baseline)
         ks_attr_test_sum = attribution_ks.detach().numpy().sum(0)
         ks_attr_test_norm_sum = ks_attr_test_sum / np.linalg.norm(ks_attr_test_sum, ord=1)
-        print("KernelShap attribution sum:", ks_attr_test_norm_sum)
         attributions_dict["KernelShap"] = ks_attr_test_norm_sum
 
     if "DeepLift" in fun_control["xai_methods"]:
         attr_dl = DeepLift(model)
         attribution_dl = attr_dl.attribute(X_val_tensor, baselines=baseline)
         dl_attr_test_sum = attribution_dl.detach().numpy().sum(0)
         dl_attr_test_norm_sum = dl_attr_test_sum / np.linalg.norm(dl_attr_test_sum, ord=1)
-        print("DeepLift attribution sum:", dl_attr_test_norm_sum)
         attributions_dict["DeepLift"] = dl_attr_test_norm_sum
 
     attributions_list = [attributions_dict[method] for method in fun_control["xai_methods"]]