Update app.py

app.py

@@ -414,7 +414,6 @@ def preview_tracks(df: pd.DataFrame, cols: list[str]):
         if i == 0: ax.set_ylabel(y_label)
         else: ax.tick_params(labelleft=False); ax.set_ylabel("")
     fig.tight_layout(); return fig
-
 # =========================
 # Load models + metas
 # =========================
@@ -446,31 +445,54 @@ def _try_load_or_explain(p: Path, name: str):
     try:
         return load_model(str(p))
     except Exception as e:
-        st.error(f"Failed to load {name} at {p}.\n\n{e}\n\nIf the error mentions `numpy._core` or versions, install the meta versions (numpy {meta_bo.get('versions',{}).get('numpy') or meta_bd.get('versions',{}).get('numpy')}, scikit-learn {meta_bo.get('versions',{}).get('scikit_learn') or meta_bd.get('versions',{}).get('scikit_learn')}).")
+        # Prefer BO meta versions if available, else BD
+        want_np  = (meta_bo.get("versions",{}) or meta_bd.get("versions",{})).get("numpy", "N/A")
+        want_skl = (meta_bo.get("versions",{}) or meta_bd.get("versions",{})).get("scikit_learn", "N/A")
+        st.error(
+            f"Failed to load {name} at {p}.\n\n{e}\n\n"
+            f"If this mentions `numpy._core` or versions, install:\n"
+            f"  • numpy {want_np}\n  • scikit-learn {want_skl}"
+        )
         st.stop()
 
-model_bo = _try_load_or_explain(bo_model_path, "BO model")
-model_bd = _try_load_or_explain(bd_model_path, "BD model")
+# Unwrap payload-style joblibs: {"model": pipeline, "model_info": {...}}
+def _unwrap(payload):
+    if isinstance(payload, dict) and "model" in payload:
+        return payload["model"], payload.get("model_info", {})
+    # Fallback: estimator saved bare or custom object
+    return payload, getattr(payload, "model_info", {})
+
+payload_bo = _try_load_or_explain(bo_model_path, "BO model")
+payload_bd = _try_load_or_explain(bd_model_path, "BD model")
 
-# Resolve features/targets strictly (no renaming)
-FEATURES = []
-if hasattr(model_bo, "feature_names_in_"):
-    FEATURES = [str(x) for x in model_bo.feature_names_in_]
-elif meta_bo.get("features"):
-    FEATURES = [str(x) for x in meta_bo["features"]]
-elif meta_bd.get("features"):
-    FEATURES = [str(x) for x in meta_bd["features"]]
-else:
-    FEATURES = FEATURES_DEFAULT[:]
+model_bo, info_bo = _unwrap(payload_bo)
+model_bd, info_bd = _unwrap(payload_bd)
 
+# Resolve features per model (prefer model_info → meta → fallback)
+features_bo = list((info_bo.get("features")
+                    or meta_bo.get("features")
+                    or FEATURES_DEFAULT))
+features_bd = list((info_bd.get("features")
+                    or meta_bd.get("features")
+                    or FEATURES_DEFAULT))
+
+# Use the UNION so BD's 'Depth (ft)' is required and both models can run from one input
+# Order: keep BO order, then append BD-only columns in order
+features_union = list(dict.fromkeys(features_bo + [c for c in features_bd if c not in features_bo]))
+
+# Resolve targets / units
 TARGET_BO = str(meta_bo.get("target") or TARGET_BO_DEFAULT)
 TARGET_BD = str(meta_bd.get("target") or TARGET_BD_DEFAULT)
+X_UNITS    = str(meta_bo.get("units") or meta_bd.get("units") or "MW (pcf)")
+
+# Stash in session
+st.session_state["FEATURES_BO"] = features_bo
+st.session_state["FEATURES_BD"] = features_bd
+st.session_state["FEATURES"]    = features_union  # strict check uses the union
+st.session_state["TARGET_BO"]   = TARGET_BO
+st.session_state["TARGET_BD"]   = TARGET_BD
+st.session_state["X_UNITS"]     = X_UNITS
 
-X_UNITS = str(meta_bo.get("units") or meta_bd.get("units") or "MW (pcf)")
-st.session_state["FEATURES"] = FEATURES
-st.session_state["TARGET_BO"] = TARGET_BO
-st.session_state["TARGET_BD"] = TARGET_BD
-st.session_state["X_UNITS"]   = X_UNITS
 
 # =========================
 # Session state