Advanced scenario parameters

Use Advanced Parameters to set more advanced parameters. Note that these parameters can also be configured in the workflow at the run configuration level. Only the Parameter value column is configurable in this table.

For more information about these parameters, see Demand forecasting overview.

Advanced Parameters

Parameter Description
Algorithm Mapping -> Local Algorithms Minimum History Specifies the minimum number of demand records pertaining to an LOD required for the LOD to be mapped to any local algorithm. If the minimum number isn't met, the LOD will instead be mapped to Stochastic Gradient Boosting (even if this algorithm wasn't specified by the user).
Algorithm Mapping -> Svm Linear Minimum History Specifies the minimum number of deamnd records pertaining ot a LOD required for the LOD to be mapped to the SVM Linear algorithm.
Algorithms -> ArimaX Hyper Parameter Dictionary of hyperparameters to be passed to the ArimaX algorithm by default.
Algorithms -> Auto Arima Hyper Parameter Dictionary of hyperparameters to be passed to the Auto Arima algorithm by default.
Algorithms -> Croston Hyper Parameter Dictionary of hyperparameters to be passed to the Croston algorithm by default.
Algorithms -> Exponential Smoothing Hyper Parameter Dictionary of hyperparameters to be passed to the Exponential Smoothing algorithm by default.
Algorithms -> Naive Hyper Parameter Dictionary of hyperparameters to be passed to the Naive algorithm by default.
Algorithms -> Prophet Hyper Parameter Dictionary of hyperparameters to be passed to the Prophet algorithm by default.
Algorithms -> Seasonal Naive Hyper Parameter Dictionary of hyperparameters to be passed to the Seasonal Naive algorithm by default.
Algorithms -> Simple Moving Average Hyper Parameter Dictionary of hyperparameters to be passed to the Simple Moving Average algorithm by default.
Algorithms -> Stochastic Gradient Boosting Hyper Parameter Dictionary of hyperparameters to be passed to the Stochastic Gradient Boosting algorithm by default.
Algorithms -> Svm Linear Hyper Parameter Dictionary of hyperparameters to be passed to the Svm Linear algorithm by default.
Backtesting Framework -> Ensemble Set Folds The number of folds to create for determining weights in ensembles. Default value is 1.
Backtesting Framework -> Estimator Selection Set Folds The number of folds to create for estimator selection. Default value is 1.
Backtesting Framework -> Fold Overlap Size Specifies the number of historical periods to share between one backtesting fold and the next. The default is 0.
Backtesting Framework -> Jump The number of periods the training set increments in each successive fold. Default value is 1.
Backtesting Framework -> Overlap Unbiased Fold Specifies whether the Fold Overlap Size (when greater than 0) applies to unbiased test folds in addition to estimator selection folds. The default is False.
Backtesting Framework -> Unbiased Test Set Folds The number of folds to create for unbiased testing.
Causal Parameters -> Include Event Info Features Specifies whether to use event info features in the models. Default value is True.
Causal Parameters -> Include Mapped Features Specifies whether to use mapped features in the models. Default value is True.
Causal Parameters -> Include Master Data Specifies whether to use the level data specified in the dim master tables. Must be set to True for Trend Cloud features to be mapped to LODs and used in models. Default value is True.
Causal Parameters -> Include Trend Cloud Features Specifies whether to use Trend Cloud features in the models. Default value is True.
Correlation Calculator -> Alter Assignments Specifies whether to subset and overwrite the initial LOD-to-feature assignments defined in the feature_assignment table based on the results of the Correlation Calculator run. Default value is True.
Correlation Calculator -> Scalability Max Features Parameter is not currently utilized by the Correlation Calculator. Default value is 200.
Correlation Calculator -> Scalability Max Lag The maximum lag of each Trend Cloud feature that will have its correlation with demand calculated by the Correlation Calculator. All tags starting with 1 up through this maximum lag is included for each Trend Cloud feature. Default value is 5.
Demand Workflow Action -> Batch Size The number of time-series included in each concurrent execution of the forecast execution action. When theparameter is left empty, the end-to-end forecast execution action does not employ a batch size, and all the time series are pocessed in a single iteration.
Effort Level -> High Algorithms The algorithms to train when the Effort Level main parameter is set to "High". If left blank, the following algorithms are trained: Naive, Seasonal Naive, Simple Moving Average, Prophet, Stochastic Gradient Boosting, ArimaX, Auto Arima, and Exponential Smoothing.
Effort Level -> Low Algorithms The algorithms to train when the Effort Level main parameter is set to "Low". If left blank, the following algorithms are trained: Naive, Seasonal Naive, Simple Moving Average, Auto Arima, and Exponential Smoothing.
Effort Level -> Medium Algorithms The algorithms to train when the Effort Level main parameter is set to "Medium". If left blank, the following algorithms are trained: Stochastic Gradient Boosting, ArimaX, Auto Arima, and Exponential Smoothing.
Ensembling -> Ensembler Weights A dictionary specifying how much weight to assign to each estimator in the ensembler (if the Forecast Strategy main parameter is set to “Ensemble”). If left blank, all estimators receive equal weight.
Estimator Selection -> Always Select Ensembler Specifies whether to always choose the ensembler over individual estimators during estimator selection when the "Forecast_Strategy" main parameter is set to "Ensemble", regardless of which estimator has the best error metric in the estimator selection folds. Default value is False.
External Features -> External Features Imputation Strategy

Specifies the method to use to impute mapped features and Trend Cloud features. Supported options are:

  • Moving Average (default)
  • Backward Fill

  • Fill NA Value

  • Forward Fill

  • Lag
External Features -> External Features Imputation Value Specifies the value to pass along with the Imputation Strategy when the strategy is Fill NA Value, Lag, or Moving Average. The default is 3.
External Features -> External Features Monthly Lag List The list of external feature lags to use as additional features in the models when time series aggregation is monthly. If left blank, the following lags are used: [1, 2, 3, 4, 11, 12, 13].
External Features -> External Features Weekly Lag List The list of external feature lags to use as additional features in the models when time series aggregation is weekly. If left blank, the following lags are used: [1, 2, 3, 4, 11, 12, 13, 51, 52, 53].
Feature Identifier -> Best Seasonal Cycles The list of seasonal cycle frequencies to apply to the Auto Arima, ArimaX, Exponential Smoothing, and Theta algorithms. Default value is 52.
Feature Identifier -> Seasonal Cycle Detection Method

The method to use to detect seasonal cycle frequencies when Best Seasonal Cycles is None. Supported options are:

  • PyCaret

  • ACF
  • FFT

PyCaret is used when this parameter is blank.

Feature Selection -> Batch Size Specifies the maximum number of historical feature matrix records to pass to each concurrent execution of the feature selection method. Default is 8000.
Feature Selection -> Max Number of Selected Features Specifies the maximum number of features that can be chosen for each time series.
Feature Selection -> Max Selected Features Percent

Limits the number of features that can be chosen for a time series to those that represent the maximum percentage of the total number of data points.

Future Forecasting -> Enable New Product Introduction? Specifies whether or not to generate forecasts for LODs that have insufficient history to be forecasted with local algorithms. Default is True.
Parallel Configuration -> Number of Cores The number of CPU cores to use. If Default, the maximum number of available CPU cores are used. The default value is Default.
Post Processing - > Round Forecast to Integer

Specifies whether to round forecasts to whole numbers. Default value is False.

Post Processing - > Trim Forecast to Zero Specifies whether to replace negative forecasts with zeros. Default value is True.
Scalability -> Max Number Of Unique Feature Sets The maximum number of unique combinations of external features to be mapped to LODs. Features are iteratively removed in ascending order of prevalence int LOD mappings until the threshold is met. Default value is 200.
Tables -> Store Estimator Configuration Specifies whether to store each estimator's arguments (hyperparameters, required_features, etc.) in the estimator_configuration table. Default value is True.
Tables -> Store Estimator Lod Mapping Specifies whether to store the list of estimator-OLD mappings that were considered (as well as indicator specifying the ones that were chosen) in the estimator_lod_mapping table. Default value is True.
Tables -> Feature Report Filter Selected Features Specifies whether or not to show only features selected for inclusion in an LOD’s local models in the Feature Report. Default value is True.
Target Time Series -> Target Time Series Include Aggregated Feature Specifies whether to use demand features aggregated by individual dimensions and subsets of dimensions in the models. Default value is False.
Target Time Series -> Target Time Series Include Rolling Mean Features Specifies whether to use demand rolling mean features in the models. Default value is True.
Target Time Series -> Target Time Series Lag List The list of demand lags to use as features in the models when time series aggregation is monthly. If left blank, the following lags are used: [1, 2, 3, 4, 11, 12, 13].
Target Time Series -> Target Time Series Rolling Mean List The list of rolling mean windows to use as features in the models when time series aggregation is monthly and "Include Rolling Mean Features?" is set to True. If left blank, the following lags are used: [3, 6, 12, 24].
Target Time Series -> Target Time Series Lag List The list of demand lags to use as features in the models when time series aggregation is weekly. If left blank, the following lags are used:  [1, 2, 3, 4, 5,11, 12, 13, 51, 52, 53].
Target Time Series -> Target Time Series Rolling Mean List The list of rolling mean windows to use as features in the models when time series aggregation is weekly and "Include Rolling Mean Features?" is set to True. If left blank, the following lags are used: [4, 8, 13, 26, 52].
Time Aggregation - > Day of Aggregation The day of the week on which weekly data aggregation occurs. The default day is Monday ('Mon').

Last modified: Friday May 12, 2023

Is this useful?