Optimize the Performance of Hyperparameter Tuning

Hyperparameter tuning refers to the ability of Oracle Demand Management to self-tune for your environment by picking machine learning parameters for your forecasted combinations that bring about the most improvement to forecast accuracy. This update includes additional capabilities, such as two more forecasting parameters, that are tuned for your environment.

This update contains these performance improvements that use background processes:

• Reduce the number of tuning iterations through a random walk algorithm.
• Shorten iteration time by disabling resource-intensive forecasting methods during hyperparameter tuning. However, the forecasting methods are enabled for the final forecast run.

When more than a few forecasting parameters are optimized, the total number of possible value permutations can get very large quickly. In this case, Demand Management switches to the random walk algorithm. Specifically, this switch occurs when the number of total permutations is greater than the sum of the value of the HypertuneMaxPermutations forecasting parameter and the number of forecasting parameter values. Otherwise, Demand Management will simply test all possible combinations.

The random walk algorithm works as follows:

1. Demand Management first tunes each forecasting parameter independently.
2. For tuning multiple forecasting parameters, Demand Management makes small, random adjustments to forecasting parameters one at a time and checks whether forecast accuracy improves.

• If the change helps, it’s retained.
• If not, the previous setting is retained.

3. This process repeats for a set number of cycles as specified by the HypertuneMaxPermutations forecasting parameter for exploring multiple possibilities.

Thus, this update results in faster hyperparameter tuning and better forecast results.

Two forecasting parameters were added for this enhancement:

• HypertuneMaxPermutations: Specifies the maximum number of permutations to be evaluated during hyperparameter tuning. If the value is 0 or 1, then all permutations are evaluated. A value greater than 1 is the maximum number of permutations evaluated for each node with the random walk algorithm. The default value is 100.
• HypertuneDisableSlowMethods: Specifies whether to disable forecasting methods that slow performance during hyperparameter tuning. Disabling these forecasting methods can significantly reduce tuning time, but they are still used during forecasting. The forecasting methods are specified by their letters. The default value is CEK, which means that the Multiplicative Monte Carlo Regression, Combined Transformation, and Multiplicative Monte Carlo Intermittent forecasting methods are disabled. When the value is null, the value is read as CEK. When the value is 0 (zero), no forecasting method is disabled.

These are the forecasting methods with their corresponding letters:

• • Auto Regressive External Inputs (X)
  • Auto Regressive Integrated External (V)
  • Auto Regressive Logistic (A)
  • Causal Winters (B)
  • Combined Transformation (E)
  • Croston for Intermittent (F)
  • Dual Group Multiplicative (D)
  • Holt (H)
  • Logistic (G)
  • Modified Ridge Regression (M)
  • Multiplicative Monte Carlo Intermittent (K)
  • Multiplicative Monte Carlo Regression (C)
  • Regression (R)
  • Regression for Intermittent (J)
  • Transformation Regression (L)
  • Naive (N)
  • Holt Naive (T)
  • Moving Average Naive (O)

Steps to Enable and Configure

You don't need to do anything to enable this feature.

Key Resources

• Refer to the Cloud Applications Readiness content for the following features for Oracle Fusion Cloud Supply Chain Planning:
  • Use Automated Hyperparameter Tuning to Improve Forecast Accuracy (24B)
  • Control Automated Hyperparameter Tuning at the Plan Level (24C)
  • Receive Planning Advisor Notifications for Automated Hyperparameter Tuning (24C)

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.