fix: adjust run_date to ensure future data exists in dataset

- Changed run_date calculation in smoke_test.py and full_inference.py
- smoke_test.py: run_date = max_date - 168 hours (7-day forecast)
- full_inference.py: run_date = max_date - 336 hours (14-day forecast)
- Ensures forecast window (Sept 17-30 or Sept 17-30) has data in dataset
- Fixes empty future_df bug that caused smoke test failure

Note: This is for smoke test validation within Sept data.
Later: proper Oct holdout validation (run_date=Sept 30, forecast Oct 1-14)

full_inference.py

@@ -81,13 +81,19 @@ print(f"     Borders: {', '.join(borders[:5])}... (showing first 5)")
 
 # Step 3: Prepare forecast parameters
 print("\n[3/7] Setting up forecast parameters...")
-run_date = df['timestamp'].max()
+# Use a date that has 14 days of future data available
+# Dataset ends at 2025-09-30 23:00, so we need run_date such that
+# forecast ends at most at 2025-09-30 23:00
+# For 336 hours (14 days), run_date should be at most 2025-09-16 23:00
 context_hours = 512
 prediction_hours = 336  # 14 days (fixed)
+max_date = df['timestamp'].max()
+run_date = max_date - timedelta(hours=prediction_hours)
 
 print(f"     Run date: {run_date}")
 print(f"     Context window: {context_hours} hours")
 print(f"     Prediction horizon: {prediction_hours} hours (14 days, D+1 to D+14)")
+print(f"     Forecast range: {run_date + timedelta(hours=1)} to {run_date + timedelta(hours=prediction_hours)}")
 
 # Initialize DynamicForecast once for all borders
 forecaster = DynamicForecast(

smoke_test.py

@@ -82,14 +82,19 @@ print(f"[*] Test border: {test_border}")
 
 # Step 3: Prepare test data with DynamicForecast
 print("\n[3/6] Preparing test data...")
-# Use last available date as forecast date (Sept 30, 23:00)
-run_date = df['timestamp'].max()
-context_hours = 512
+# Use a date that has 7 days of future data available
+# Dataset ends at 2025-09-30 23:00, so we need run_date such that
+# forecast ends at most at 2025-09-30 23:00
+# For 168 hours (7 days), run_date should be at most 2025-09-23 23:00
 prediction_hours = 168  # 7 days
+max_date = df['timestamp'].max()
+run_date = max_date - timedelta(hours=prediction_hours)
+context_hours = 512
 
 print(f"     Run date: {run_date}")
 print(f"     Context: {context_hours} hours (historical)")
 print(f"     Forecast: {prediction_hours} hours (7 days, D+1 to D+7)")
+print(f"     Forecast range: {run_date + timedelta(hours=1)} to {run_date + timedelta(hours=prediction_hours)}")
 
 # Initialize DynamicForecast
 forecaster = DynamicForecast(