Upd rolling features

app.py

@@ -153,7 +153,7 @@ def ingest(entry: OBDEntry, background_tasks: BackgroundTasks):
     if entry.status == "end":
         background_tasks.add_task(process_data, norm_ts)
         return {"status": "processed"}
-    # Normal row append
+    # Moderate row append
     try:
         df = pd.read_csv(RAW_CSV)
         row = {"timestamp": norm_ts, "driving_style": entry.driving_style}
@@ -477,8 +477,8 @@ def _process_and_save(df, norm_ts):
         import traceback
         logger.error(f"❌ UL labeling traceback: {traceback.format_exc()}")
         # Fallback: provide default labels
-        logger.warning("⚠️ Using fallback: default 'Normal' labels")
-        DRIVE_STYLE = ["Normal"] * len(df)
+        logger.warning("⚠️ Using fallback: default 'Moderate' labels")
+        DRIVE_STYLE = ["Moderate"] * len(df)
         df_labeled = df.copy()
         df_labeled["driving_style"] = DRIVE_STYLE
         labeled_path = os.path.join(CLEANED_DIR, f"cleaned_{norm_ts}_labeled.csv")
@@ -492,7 +492,9 @@ def _process_and_save(df, norm_ts):
         efficiency_labeler = EfficiencyLabeler.get()
         efficiency_preds = efficiency_labeler.predict_df(df)
         # update global FUEL_EFFICIENCY (overwrite if already exists)
-        FUEL_EFFICIENCY = [float(p) for p in efficiency_preds]
+        # Efficiency is predicted per drive, so replicate the single score for all rows
+        efficiency_score = float(efficiency_preds[0]) if efficiency_preds else 0.0
+        FUEL_EFFICIENCY = [efficiency_score] * len(df_for_persist)
         # write efficiency CSV (fuel_efficiency column)
         df_efficiency = df_for_persist.copy()
         df_efficiency["fuel_efficiency"] = FUEL_EFFICIENCY

utils/dbehavior_labeler.py

@@ -128,8 +128,8 @@ def engineer_features(df):
     def add_roll(col):
         if col not in fe.columns: 
             return
-        # Safety check to prevent infinite recursion
-        if col.startswith(('SPEED_', 'RPM_', 'THROTTLE_', 'MAF_', 'ENGINE_', 'ACCEL_', 'JERK_')):
+        # Safety check to prevent infinite recursion - only skip if it already has rolling suffix
+        if any(col.endswith(suffix) for suffix in ['_mean_w1', '_std_w1', '_mean_w2', '_std_w2', '_mean_w5', '_std_w5', '_mean_w8', '_std_w8', '_min_w1', '_max_w1']):
             return  # Skip if already a rolling feature
         for w, tag in [(W1, "w1"), (W2, "w2"), (W5, "w5"), (W8, "w8")]:
             fe[f"{col}_mean_{tag}"] = fe[col].rolling(w, min_periods=1, center=True).mean()