Update backend/core/explainability.py

backend/core/explainability.py

@@ -1,33 +1,48 @@
-import shap
-import numpy as np
-
-class ExplainabilityEngine:
-    def explain_tabular(self, model_pipeline, X_sample):
-        if X_sample.empty:
-            raise ValueError("Sample data is empty, cannot compute explanations")
-
-        # Extract trained model and preprocessor
-        preprocessor = model_pipeline.named_steps["preprocessor"]
-        model = model_pipeline.named_steps["model"]
-
-        X_transformed = preprocessor.transform(X_sample)
-
-        if X_transformed.shape[0] == 0:
-            raise ValueError("Transformed sample data is empty after preprocessing")
-
-        explainer = shap.Explainer(model, X_transformed)
-        shap_values = explainer(X_transformed, check_additivity=False)
-
-        if shap_values is None or shap_values.values is None:
-            raise ValueError("SHAP computation failed")
-
-        global_importance = np.abs(shap_values.values).mean(axis=0).tolist()
-
-        if len(global_importance) == 0:
-            raise ValueError("No feature importance computed")
-
-        return global_importance
-
-
-
-
+import shap
+import numpy as np
+from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
+from sklearn.linear_model import LinearRegression, LogisticRegression
+
+
+class ExplainabilityEngine:
+    def explain_tabular(self, model_pipeline, X_sample):
+        if X_sample.empty:
+            raise ValueError("Sample data is empty, cannot compute explanations")
+
+        preprocessor = model_pipeline.named_steps["preprocessor"]
+        model = model_pipeline.named_steps["model"]
+
+        X_transformed = preprocessor.transform(X_sample)
+
+        if X_transformed.shape[0] == 0:
+            raise ValueError("Transformed sample data is empty after preprocessing")
+
+        # -------- Model-aware SHAP selection --------
+        if isinstance(model, (RandomForestClassifier, RandomForestRegressor)):
+            explainer = shap.TreeExplainer(model)
+            shap_values = explainer.shap_values(
+                X_transformed, check_additivity=False
+            )
+
+            # Classification returns list
+            if isinstance(shap_values, list):
+                shap_values = shap_values[1]
+
+        elif isinstance(model, (LinearRegression, LogisticRegression)):
+            explainer = shap.LinearExplainer(model, X_transformed)
+            shap_values = explainer.shap_values(X_transformed)
+
+        else:
+            raise ValueError(
+                f"Explainability not supported for model type: {type(model)}"
+            )
+
+        if shap_values is None or len(shap_values) == 0:
+            raise ValueError("SHAP computation failed")
+
+        global_importance = np.abs(shap_values).mean(axis=0).tolist()
+
+        if not global_importance:
+            raise ValueError("No feature importance computed")
+
+        return global_importance