Update src/streamlit_app.py

src/streamlit_app.py

@@ -331,34 +331,46 @@ df, meta_df = load_data()
 
 
 # -------------------------
-# Sidebar filters & UI (fault-tolerant for minimal metadata)
+# Sidebar filters & UI (FINAL ROBUST VERSION)
 # -------------------------
 st.sidebar.title("Feature Explorer - Advanced + SHAP")
 
-# Ensure meta_df always has at least placeholder columns
-required_cols = ["feature_name", "source_type", "formula", "remarks"]
-for col in required_cols:
-    if col not in meta_df.columns:
-        meta_df[col] = None
-
-
-# Populate placeholders if metadata is summary-like (not feature-level)
-if len(meta_df) < len(df.columns) or "feature_name" not in meta_df.columns:
-    st.warning("Metadata appears to be summary-only. Rebuilding feature-level metadata.")
-    meta_df = pd.DataFrame({
-        "feature_name": df.columns,
-        "source_type": [
-            "engineered" if any(x in c for x in ["poly", "pca", "roll", "lag"]) else "measured"
-            for c in df.columns
-        ],
-        "formula": ["" for _ in df.columns],
-        "remarks": ["auto-inferred synthetic feature metadata" for _ in df.columns]
-    })
+def ensure_feature_metadata(df: pd.DataFrame, meta_df: pd.DataFrame) -> pd.DataFrame:
+    """Ensure metadata dataframe matches feature count & has required columns."""
+    required_cols = ["feature_name", "source_type", "formula", "remarks"]
+
+    # If metadata missing or too short, rebuild it entirely
+    if meta_df is None or len(meta_df) < len(df.columns):
+        meta_df = pd.DataFrame({
+            "feature_name": df.columns,
+            "source_type": [
+                "engineered" if any(x in c for x in ["poly", "pca", "roll", "lag"]) else "measured"
+                for c in df.columns
+            ],
+            "formula": ["" for _ in df.columns],
+            "remarks": ["auto-inferred synthetic feature metadata" for _ in df.columns],
+        })
+        st.sidebar.warning("Metadata was summary-only — rebuilt feature-level metadata.")
+    else:
+        # Ensure required columns exist
+        for col in required_cols:
+            if col not in meta_df.columns:
+                meta_df[col] = None
+        # Fill feature_name if blank or NaN
+        if meta_df["feature_name"].isna().all():
+            meta_df["feature_name"] = df.columns
+        # Clip to same number of features (safety)
+        if len(meta_df) > len(df.columns):
+            meta_df = meta_df.iloc[: len(df.columns)]
+
+    return meta_df
 
+meta_df = ensure_feature_metadata(df, meta_df)
 
 # Build sidebar safely
 feat_types = sorted(meta_df["source_type"].dropna().unique().tolist())
 selected_types = st.sidebar.multiselect("Feature type", feat_types, default=feat_types)
+
 if "source_type" not in meta_df.columns or meta_df["source_type"].dropna().empty:
     filtered_meta = meta_df.copy()
 else:
@@ -366,6 +378,7 @@ else:
 
 numeric_cols = df.select_dtypes(include=[np.number]).columns.tolist()
 
+
 # -------------------------
 # Features tab (robust)
 # -------------------------