Forecast engine warning codes
This table list the warning messages that can be returned when generating a forecast. These are
for informational purpose and for additional review. In spite of warning, the engine returns a
forecast, but the results can differ from the original forecast engine configuration.
| Code | Warning message | Algorithm or Parameter (if applicable) | Description |
|---|---|---|---|
| 1 | There is an event with no history. | Events only algorithm | The forecast engine is using Events Only algorithm. However, there is a historical period which contains an event value with no corresponding history value. A value of ‘0’ is assumed for this event for calculating of average event size. |
| 2 | There is history with no event. | Events Only algorithm | The forecast engine is using Events Only algorithm. However, there is a historical period which contains an event value with no corresponding history value. The history point value is not used for calculating the average event size. |
| 3 | Forecast combining has been selected and fails the forecast magnitude test. | Best algorithm with Forecast Combining. Forecast Test Magnitude |
The Best Fit selection returned a combined forecast for this item. However, the forecast magnitude test is failed. A Review Alerts (Forecast Magnitude Error) is returned for this item-location to review the Forecast Test Magnitude parameter. |
| 4 | A linear/seasonal Holt-Winters model could not be fitted using the AVERAGING initialisation as the history length after the initialisation period has been removed is 0 or 1. An alternative model has been used instead. | Best algorithm | The history length of item matches or is 1 period longer than the forecast engine Periodicity. There is not enough historical data available to perform a Holt-Winters linear or seasonal model using Averaging initialization, hence these model types are skipped during the Best Fit process. |
| 5 | There is at least one forecast event type that has no corresponding history event type. | Events Only algorithm or Perform Event Modeling | At least one event profile indicates a future event for which there is no corresponding historical event type is available. This means no average event size is calculated and hence the forecasts event value cannot be calculated. As a result of this, the baseline forecast value is applied to that forecast period. |
| 6 | There are not enough non-event periods to calculate average event size based on Minimum Events Length {0}. All events have been removed and a model has been fit to the unaltered history. | Event Modeling Method = LEASTSQUARES | The forecast engine is set to Event Modeling Method = LeastSquares. However, the item does not have enough historical periods without a historical event to create a baseline forecast (without events). The item-location has been forecast as a normal, or non-event item. |
| 7 | The history replacement method has been changed to Naïve as there is not enough history for the given method. | History Replacement. Algorithm | The engine is set to perform history replacement, but the item-location combination does not have enough history data to process using the defined algorithm. Review the history data of the item-location or consider an alternative Algorithm. |
| 10 | There is not enough history to perform the extended Shapiro-Wilks normality test on the residuals. No test has been performed. | Shapiro-Wilks Residual Analysis | The Shapiro-Wilks test could not be performed due to a lack of history and therefore this specific residual analysis could not be determined. |
| 11 | The calculated maximum lag is too low due to the low number of residuals. The Ljung-Box test for autocorrelation in the residuals has failed. | Ljung-Box Residual Analysis | The Ljung-Box test for autocorrelation could not be performed due to a low number of residual values, therefore this specific residual analysis could not be determined. |
| 12 | The degrees of freedom for the F distribution are too near or below zero, that is, the number of residuals is too low. The F test for constant variance in the residuals has failed. | F-test Residual Analysis | The F-test for constant variance could not be performed, therefore this specific residual analysis could be determined. |
| 13 | The degrees of freedom for the T distribution are too near or below zero, that is, the number of residuals is too low. The t test for zero mean in the residuals has failed. | One-sample t Residual Analysis | The One-sample t statistic for a zero mean could not be performed, therefore this specific residual analysis could not be determined. |
| 14 | Ljung-Box Q Statistic P Value cannot be calculated as the history ({0}) and online forecast ({1}) are of different lengths. | Ljung-Box Residual Analysis | The Ljung-Box test for autocorrelation could not be performed due to differing history and online forecast lengths, therefore this specific residual analysis could not be determined. |
| 15 | The calculation of the auto correlation coefficient is not possible because the lag value ({0}) is greater than or equal to the data length ({1}). | Ljung-Box Residual Analysis | The Ljung-Box test for auto-correlation could not be performed, therefore this specific residual analysis could not be determined. |
| 16 | There has been an error in the use of the Normal distribution function. | Shapiro-Wilks Residual Analysis | The Shapiro-Wilks test could not be performed, therefore this specific residual analysis could not be determined. |
| 17 | There are too many data points to reasonably calculate the p value. | Shapiro-Wilks Residual Analysis | The Shapiro-Wilks test could not be performed due to too many data points, therefore this specific residual analysis could not be determined. |
| 18 | History Events Exist with No Forecast Events | Event Only algorithm, Event Modeling Method | The engine is set to perform event modeling and the item-location combination has some historical events, however no future events are defined to add the effects of events in to the forecast. The Average Event Size and baseline forecast without events are still calculated, however a warning is raised. If this is a valid situation for this item-location, the warning can be ignored, else review the data for future periods of the measures mapped to event types in the forecast engine. |
| 19 | There has been a failure in the BATS baseline method for event modelling. The Least Squares method has been used. | Event Modeling Method = Dynamic BATS | The engine is set to perform Event Modeling Method with Dynamic BATS, however there may be cases where the item being forecast tries to use BATS for baselining. This fails and the engine uses Least Squares instead. |