Fix age range handling and model architecture

inference_gohan_cid.py

@@ -55,7 +55,7 @@ class GohanCIDInferenceEngine:
 			return None
 		# Use training-script hyperparameters
 		model = FTTransformer.make_baseline(
-			n_num_features=8,
+			n_num_features=5,  # Updated: 5 numerical features (age ranges are now categorical)
 			cat_cardinalities=self.cat_cardinalities,
 			d_out=len(self.all_cids),
 			d_token=768,  # Use the actual saved model's d_token
@@ -103,9 +103,24 @@ class GohanCIDInferenceEngine:
 				X_cat.append(self._encode_categorical(self.cat_encoders[col], '__UNKNOWN__'))
 		X_cat = torch.tensor([X_cat], dtype=torch.long)
 
-		# Numerical features (8 features to match training script)
-		num_cols = ['LAT', 'LONG', 'DELIVERY_NUM', 'MEDIAN_GENDER_RATIO', 'MODE_TOP_AGE_RANGE_1', 'MODE_TOP_AGE_RANGE_2', 'MODE_TOP_AGE_RANGE_3', 'TOTAL_VOLUME']
-		X_num = torch.tensor([[float(df.get(c, pd.Series([0])).iloc[0]) for c in num_cols]], dtype=torch.float32)
+		# Numerical features (5 features to match training script - age ranges are now categorical)
+		# Remove age range fields from numerical features since they're now categorical
+		num_cols = ['LAT', 'LONG', 'DELIVERY_NUM', 'MEDIAN_GENDER_RATIO', 'TOTAL_VOLUME']
+		X_num = []
+		for col in num_cols:
+			if col in df.columns:
+				try:
+					X_num.append(float(df[col].iloc[0]))
+				except (ValueError, TypeError):
+					X_num.append(0.0)
+			else:
+				# Provide default values for missing fields
+				if col == 'TOTAL_VOLUME':
+					X_num.append(0.0)  # Default total volume
+				else:
+					X_num.append(0.0)
+		
+		X_num = torch.tensor([X_num], dtype=torch.float32)
 		return X_cat, X_num
 
 	def predict(self, data: Dict) -> List[Dict]: