How You Improve the Performance of Hyperparameter Tuning

This topic discusses ways in which you can improve the performance of hyperparameter tuning for the forecasting parameters for your user-defined forecasting profile of the Bayesian machine-learning type for your demand or demand and supply plan.

Use these forecasting parameters in the Hyperparameter Tuning category to improve the performance of hyperparameter tuning:

• HypertuneMaxPermutations: Use this forecasting parameter to specify the maximum number of permutations of hyperparameter-tuning values that should be evaluated.

  The hyperparameter-tuning values are those that are provided in the Hyperparameter Tuning Values column in the Parameters step in the guided process for creating or editing a forecasting profile. If no values are available in the column, the default values in the HypertuneParamSet forecasting parameter are used for selection of the best values.

  If you enter 0 (zero) or 1, all permutations are evaluated for a forecasting node (combination). Any value greater than 1 is the maximum number of permutations evaluated for the forecasting node through a random-walk algorithm. The default value is 100.

  The random-walk algorithm is used when the number of permutations is greater than the sum of the value of the HypertuneMaxPermutations forecasting parameter and the number of values for hyperparameter tuning. Otherwise, the algorithm isn't used, and all combinations for the hyperparameter-tuning values are evaluated for the forecasting node.

  The random-walk algorithm works as follows:

  1. Each forecasting parameter with hyperparameter-tuning values is evaluated independently.
  2. For all the forecasting parameters with hyperparameter-tuning values taken together, small, random changes are made to the values one at a time, and the forecast accuracy is evaluated.
     • If the forecast accuracy improves, the change is accepted.
     • If the forecast accuracy doesn't improve, the previous value is retained.

  These steps are repeated for each forecasting node for the number of times specified by the value of the HypertuneMaxPermutations forecasting parameter.

• HypertuneDisableSlowMethods: Use this forecasting parameter to disable forecasting methods during hyperparameter tuning. By disabling some forecasting methods, you can significantly reduce the time required for hyperparameter tuning. However, the forecasting methods that you disable are still used during the calculation of the final forecast for forecasting nodes.

  You can specify the forecasting methods by their letters. The default value is CEK, which means that the forecasting methods named Multiplicative Monte Carlo Regression (C), Combined Transformation (E), and Multiplicative Monte Carlo Intermittent (K) are disabled. If the value of the forecasting parameter is null, the value is considered as CEK. If you enter 0 (zero) as the forecasting-parameter value, no forecasting method is disabled.

  For information about the forecasting methods and their letters, see the topic titled Forecasting Methods.