Time Series to Model (Temporal Causal Modeling)
On the Fields tab, use the Time Series settings to specify the series to include in the model system.
For column-based data, the term series has the same meaning as the term field. For multidimensional data, fields that contain time series are referred to as metric fields. A time series, for multidimensional data, is defined by a metric field and a value for each of the dimension fields. The following considerations apply to both column-based and multidimensional data.
- Series that are specified as candidate inputs or as both target and input are considered for inclusion in the model for each target. The model for each target always includes lagged values of the target itself.
- Series that are specified as forced inputs are always included in the model for each target.
- At least one series must be specified as either a target or as both target and input.
- When Use predefined roles is selected, fields that have a role of Input are set as candidate inputs. No predefined role maps to a forced input. For more information, see the topic Roles.
Multidimensional data
For multidimensional data, you specify metric fields and associated roles in a grid, where each row in the grid specifies a single metric and role. By default, the model system includes series for all combinations of the dimension fields for each row in the grid. For example, if there are dimensions for region and brand then, by default, specifying the metric sales as a target means that there is a separate sales target series for each combination of region and brand.
For each row in the grid, you can customize the set of values for any of the dimension fields by clicking the ellipsis button for a dimension. This action opens the Select Dimension Values subdialog. You can also add, delete, or copy grid rows.
The Series Count column displays the number of sets of dimension values that are currently specified for the associated metric. The displayed value can be larger than the actual number of series (one series per set). This condition occurs when some of the specified combinations of dimension values do not correspond to series contained by the associated metric.