Improve AI logic

services/training/app/services/training_orchestrator.py

@@ -170,6 +170,7 @@ class TrainingDataOrchestrator:
             logger.error(f"Training data preparation failed: {str(e)}")
             raise ValueError(f"Failed to prepare training data: {str(e)}")
     
+    @staticmethod
     def extract_sales_date_range_utc_localize(sales_data_df: pd.DataFrame):
         """
         Extracts the UTC-aware date range from a sales DataFrame using tz_localize.
@@ -246,12 +247,14 @@ class TrainingDataOrchestrator:
                 if 'date' in record:
                     record_date = record['date']
                     
-                    # ✅ FIX: Proper timezone handling for date parsing
+                    # ✅ FIX: Proper timezone handling for date parsing - FIXED THE TRUNCATION ISSUE
                     if isinstance(record_date, str):
+                        # Parse complete ISO datetime string with timezone info intact
+                        # DO NOT truncate to date part only - this was causing the filtering issue
                         if 'T' in record_date:
                             record_date = record_date.replace('Z', '+00:00')
-                        # Parse with timezone info intact
-                        parsed_date = datetime.fromisoformat(record_date.split('T')[0])
+                        # Parse with FULL datetime info, not just date part
+                        parsed_date = datetime.fromisoformat(record_date)
                         # Ensure timezone-aware
                         if parsed_date.tzinfo is None:
                             parsed_date = parsed_date.replace(tzinfo=timezone.utc)
@@ -260,8 +263,8 @@ class TrainingDataOrchestrator:
                         # Ensure timezone-aware
                         if record_date.tzinfo is None:
                             record_date = record_date.replace(tzinfo=timezone.utc)
-                        # Normalize to start of day
-                        record_date = record_date.replace(hour=0, minute=0, second=0, microsecond=0)
+                        # DO NOT normalize to start of day - keep actual datetime for proper filtering
+                        # Only normalize if needed for daily aggregation, but preserve original for filtering
                     
                     # ✅ FIX: Ensure aligned_range dates are also timezone-aware for comparison
                     aligned_start = aligned_range.start
@@ -885,4 +888,4 @@ class TrainingDataOrchestrator:
                     1 if len(dataset.traffic_data) > 0 else 0
                 ])
             }
-        }
+        }