Overview of generating forecasts at a non-base period level
You can call the forecast engine, to fit a model at a period level that does not match the Cycle.Calendar level to store the scenario values.
Aggregation of periods
If aggregation is required for the defined periods, you must aggregate the values for all base-level periods, within each Month, for the Cycle.History Horizon. These are the applicable measures for aggregation:
- Forecast Engine.Measure Mapping.History
- Forecast Engine.Measure Mapping.Weighting
- Forecast Engine.Measure
Mapping.Mask Note: This measure is not used for a direct aggregation of values. The mask (data used for bitwise operations such as AND or NOT) must applied for all base-level periods that are masked (same period mask value > 0). This ensures that the history measure for that base-level period is not to the period level (parent). If all base-level periods for the same parent have a mask > 0, then the aggregated period mask, to be passed to the forecast engine is 1.
When generating forecasts at a non-base period level, aggregration is not required for Growth Damping input measures (growthDampingStartPeriod, growthDampingEndPeriod, growthDampingFactor, and gdfZeroAfterDamping) as the values are stored at the 'pconst' level.
For example, to aggregate weeks to months, if the weeks 1-4 have an aggregate month equal to 1 and the history for weeks 1-4: 100, 100, 100, 100. Then, the mask for weeks for 1-4 are 1,0,0,0 respectively. The aggregated history for month 1 is 300, and the aggregated mask for month 1 is 0.
To generate an aggregate, the forecast engine requires the complete history for the month. So, the number of periods at Cycle.Calendar to store scenario values <= Cycle.History Horizon must be used to ensure that valid Monthly periods are passed to the Forecast Engine. For example, a history horizon comprises of 14 weeks, which is 3 monthly periods, using the 445 weekly pattern. Therefore, the fourth monthly period cannot be calculated, so the 14th week is ignored; unless the specified weekly pattern is 455.