Compute Best Prediction from all Models

This node will search through the global dictionary for deployment information computed for the participating models. If more than one model was used to compute predictions, an average response will be compute by default. Options for voting (bagging) are also provided.

Note that the output spreadsheet will contain only those cases selected by the case selection conditions defined for the InputDescriptor marked for deployment, and that case weights will be ignored.

Parameter 1

Method for final prediction. Specifies the method for computing the final predicted value, from the currently available trained models (prediction methods).

Subset input variables. Specifies whether to carry along all variables for the generated data source with predicted values; if you select No, then only those variables selected for the analysis (as dependent variables or predictors) will be copied into the results spreadsheet (data source) for further analyses.

Selection of variables for deployment. Selects how the variables (predictors and dependent variable) for the analysis (computation of prediction or predicted classifications) are to be determined.

If you choose From PMML file, then any existing (user-defined) variable selections will be ignored, and the program will attempt to automatically determine the variables in the data file that match those (using the same name) referenced as predictors or dependent variable in the PMML file.

If you select User defined (selected) then any references to specific variables in the PMML file will be ignored, and the model specified in the PMML code will be applied to the variables selected by the user.