Receive Planning Advisor Notifications for Automated Hyperparameter Tuning

Before this update, you could use automated hyperparameter tuning to find the best values for various forecasting parameters to improve forecast accuracy in your demand plan. With this update, you can choose to be notified through the Planning Advisor of the availability of tuned forecast settings, including the improvement to the computed forecast accuracy (mean absolute percentage error (MAPE)), for the sampled data. From the Planning Advisor, you can link to a page layout that you can configure.

A recommendation type called Summary of hyperparameter tuning has been introduced in the Planning Advisor to notify users when hyperparameter tuning improved the forecast for one or more nodes during the plan run.

Planning Advisor with Collapsed Details of Hyperparameter Tuning

Planning Advisor with Collapsed Details of Hyperparameter Tuning

The Planning Advisor displays summary information on the number of nodes that tuning was performed on and the resulting improvement.

Planning Advisor with Expanded Details of Hyperparameter Tuning

Planning Advisor with Expanded Details of Hyperparameter Tuning

The following are explanations of some details:

  • These are the cumulative results of hyperparameter tuning from 16/05/2024 to 16/05/2024.
    • The first date displays the date of the first run of the forecasting profile with hyperparameter tuning. The second date displays the date of the last run of the forecasting profile with hyperparameter tuning. In this example, we had the runs on the same date.
  • 9 nodes were evaluated.
    • This number represents the nodes that were evaluated in the runs.
  • Improvements to the forecast accuracy were identified for 3 nodes (33.33% of the evaluated nodes).
    • 3: This number represents nodes for which the forecasting parameters were successfully tuned.
    • 33.33: This number represents the percentage of the evaluated nodes for which the forecasting parameters were successfully tuned.
  • The MAPE improvement for the tuned nodes was 1.5% (weighted average at the forecast level). The reduction in the mean absolute error was 1780.89 units.
    • 1.5: This number represents the improvement to the MAPE as a percentage value (weighted average).

The calculation for this value is as follows:

SUM((<output measure name>Tuned Base MAPE – <output measure name>Tuned Best MAPE)*InputMeasureHistAvg*ForecastBuckets)

Where <output measure name>Tuned Base MAPE and <output measure name>Tuned Best MAPE are measures, InputMeasureHistAvg is the average demand over the length of history that’s used to drive proportions, and ForecastBuckets is the number of forecast buckets for the plan

    • 1780.89: This number represents the improvement to the MAPE in units.

The calculation for this value is as follows:

SUM((<output measure name>Tuned Base MAPE – <output measure name>Tuned Best MAPE)*InputMeasureHistAvg)/SUM(InputMeasureHistAvg)

Where <output measure name>Tuned Base MAPE and <output measure name>Tuned Best MAPE are measures, and InputMeasureHistAvg is the average demand over the length of history that’s used to drive proportions

Steps to Enable

No steps are required to enable this feature.

Access Requirements

Users who are assigned a configured job role that contains these privileges can access this feature:

  • Edit Forecasting Profiles (MSC_EDIT_FORECASTING_PROFILES_PRIV)
  • Edit Plan Options (MSC_EDIT_PLAN_OPTIONS_PRIV)
  • Edit Plans (MSC_EDIT_PLANS_PRIV)

These privileges were available prior to this update.