Forecasting tutorial: data spreading methods for forecasts
Before you begin
This tutorial is meant to be followed in sequential order. Each part assumes that the previous part was completed.
This tutorial uses the pa_data_spreading_demo.csv sample data file, which you downloaded in Forecasting tutorial: preparing data
Procedure
- Create a new Planning Analytics Workspace book.
- Expand the 24Retail database on the Data tree.
- Right-click Cubes, then click Import
data.
- On the Import data window, click New cube.
- Name the cube Spreading_<your initials>, using your initials as a suffix.
- Upload the pa_data_spreading_demo.csvfile, then click Next.
- Examine the preview of the pa_data_spreading_demo.csv file, then
click Next. Note that the file contain two columns that contain geographic information. You'll set mapping options to use these two columns to create consolidations in a Region dimension.
- Ensure that the Parent-Child mapping format option is selected on the Map data panel.
- Click Overwrite existing values.
- Click Save process as, then enter
Spreading_data_import_<your initials> as the process name.
Now it's time to map the columns in the pa_data_spreading_demo.csv file to Planning Analytics dimensions. This is done in the Data source mapping section of the Map data panel.
- Click Add dimension. The source data has columns that represent three cube dimensions, but the default mapping accommodates only two dimensions. You must add the third dimension.
- Click the first Enter dimension name cell and enter Product_<your initials>.
- Click the corresponding Select mapping cell on the Child column, then click Products_sample.
- Click the next Enter dimension name cell and enter Time_<your initials>.
- Click the corresponding Select mapping cell on the Child column, then click Time_sample.
- Click the final Enter dimension name cell and enter Region_<your initials>.
- Click the corresponding Select mapping cell on the Parent column, then click Country.
- Click the corresponding Select mapping cell on the
Child column, then click City.
The last column in the data source is correctly identified as Data of Numeric type, but it isn't yet mapped.
- Click the Select mapping cell on the
Child column that corresponds to Data, then click
Value.
The Map data panel should now look like this:
- Click Import.
- Click OK after you review the import details. The Spreading_<your initials> cube and its associated dimensions are created on the 24retail database. The default view of the cube opens in a new book.
- Drag the dimension tiles to configure a view with Product_<your
initials>and Region_<your initials> dimensions stacked on the
rows and Time_<your initials> on the columns.
Review the data in the view and note that there are no glove sales in Miami for any time period.
- Select the Gloves, USA row, then click the
Forecast icon
on the Exploration toolbar.
- On the Forecast panel, set the Forecast period start to Jan-23. The Forecast period end should automatically be identified as Dec-23.
- Click Preview to preview the forecast. In this exercise, do not correct any outliers.
- Dismiss the preview, then click the Advanced tab on the Forecast
panel. Notice that by default, the forecast operation uses the Proportional
data spreading method to forecast values.
- Click Forecast. Examine the forecast values and note that the Proportional data spreading method creates forecast values for gloves in Miami, while there were no glove sales in that location prior to the forecast period.
This is because consolidations in the time series (in this case, USA) are forecast using proportional spreading, but values for members of the consolidations are created by equally spreading the consolidation value to all members. In this example the result is unsatisfactory because historically there have been no glove sales in Miami; the forecast should not predict future glove sale.
Fortunately, you can correct this forecast by changing the forecast data spreading method from Proportional to Relative Proportional.
- Click the cell at the intersection of Gloves, USA and
Jan-23, then enter 0> to clear the results of the
forecast.
- If the Forecast panel is not visible, click
.
- Click the Advanced tab on the Forecast panel.
- Click Edit on the Spread forecast values section of the Advanced
tab.
- Click Relative proportional on the Spread data options dialog.
- Select Range of time periods from the Reference
Type list.
You can define the time period you want to use as the reference values for the proportional data spread. In this case, you want to spread forecast values relative to the values for the same region and product in the previous year.
- From the Start of time period list, select Jan-22.
- From the End of time period list, select
Dec-22.
- Click Apply.
- Click Forecast. Because the forecast now spreads values proportional to existing values in a defined range, there are no forecast values for gloves in Miami, as there were no values for that region and product in the reference range.