To use the Amazon Ensemble algorithm, you need to create a model-level explainability module and import action.

These prerequisites should be in place:

  • Your model must have historic and related data export actions defined. 
  • You must run a data collection successfully.
  • Data for this module depends on a forecast model trained with Amazon Ensemble.

To create a model-level explainability module and import action:

  1. Create a list called "Forecast Models."
  2. In Pivot view:
    • Place Forecast Feature Names as Rows dimensions.
    • Place Line Items as the Column dimension and remove Time.
    • Place the Forecast Models in the Pages dimension (see screenshot 1 below).
    • Select OK.
Pivot view of the setup for model-level explainability.
  1. Add a line item called Feature Contribution.
  2. In Blueprint view, select None in the Summary column for the Feature Contribution line item.
null
  1. From the regular view:
    • Select Import from the Data dropdown.
  2. From the Select Source dialog:
    • Select import_action_setup_model_explainability.csv.
    • Click the Select button.
  3. From the File Options dialog:
    • Use the Set Default File dropdown to select Admins Only.
    • Select Next.
  4. From the Import dialog, Mapping tab:
    • Map Source to Target. The table below provides an example. Your sources and targets will vary.
SourceTarget
Column 4 FORECAST_MODEL_NAMEForecast Model
Column 1 FEATURE_NAMEList Forecast Feature Names
(Column Headers)Model-Level Explainability Module
    • Select Only update imported cells on the right (see screenshot 2 below).
    • Note: You don't need to map within the Forecast Feature name tabs. The Forecast Feature name tab will have no Target items until you run a Forecast action.
    • From the Forecast Explainability Line Items tab:
Source itemsMapped To
FEATURE_CONTRIBUTIONFeature Contribution
    • Select Run Import.
      This is how you define the import to the explainability module.
    • Optional: from Import actions, rename import_action_setup_explainability to IMP Module Forecast Explainability.

Blueprint setup example:

Mapping selections for Forecast of Model-level explainability, the import action setup.

9. Select OK.
Your explainability module is now ready. The next step is to import your model-level explainability.


Disclaimer

We update Anapedia regularly to provide the most up-to-date instructions.