Select Output

The Select Output page lets you define what type of predictive metrics are created for the training metric. You have the following options:

• Predictive metric base name: Type a name that serves as a default base name for any predictive metric that is created from the training metric. This base name is then combined with the name of the predictor types you select to create a full name for the predictive metrics. You can modify these names directly in the Name column of the Predictive Metric(s) to Generate area.

• Destination: Select the destination of where each predictive metric is saved. Your options depend on the type of training metric being created:

• Folder: This option is available if you are creating stand-alone training metrics or derived training metrics. Click... (the Browse button) to define the location in your MicroStrategy project to save the predictive metrics created by the training metric. If you select this option, predictive metrics are saved as stand-alone predictive metrics.
  
  If you are using derived training metrics, you should select this option when a predictive model worthy of deployment is found. Selecting this option saves the predictive model as stand-alone predictive metrics.

• Report Objects: This option is available if you are creating derived training metrics only. Select this option to create the predictive metrics as derived predictive metrics. A derived predictive metric exists only within the report used to create the predictive metric. This capability is particularly useful during the exploratory phase of the data mining process, during which you can test many variables and create a large variety of models.

The way in which derived predictive metrics are created depends on whether the derived training metrics are included directly on a report or on a report included in a document as a Grid/Graph:

• When the report is executed to create predictive metrics, the following occurs:

  • The derived training metrics are moved off the report's template; the template determines what is displayed for the report. The derived training metrics are still available in the Report Objects of the report, which supplies the definition of the report. If you update the derived training metrics and need to re-create their associated derived predictive metrics, you must add the derived training metrics back to the template of the report. When the derived predictive metrics are re-created, new versions are created to keep a history of the predictive modeling.

  • Any derived training metrics that are moved off the report's template during report execution are replaced with the first predictive metric selected to be generated for the derived training metric. When the derived training metric is initially replaced with the derived predictive metric, the values are not automatically updated to reflect the predictive metric analysis. Re-execute the report to update the values for the derived predictive metric.

  • All other predictive metrics selected to be generated are added to the Report Objects and are not on the report grid by default. This makes the predictive model available for viewing only. To execute the derived predictive metric, you must add the derived predictive metric to the grid of the report.

• When the document including the report as a Grid/Graph is executed to create predictive metrics, the following occurs:

  • All predictive metrics are added to the available dataset objects for the report and are not included on the report grid by default. These predictive metrics are not saved in the report itself and are only available with the report as a dataset in the current document. This makes the predictive model available for viewing only. To execute the derived predictive metrics, you must add the derived predictive metrics to the grid of the report or directly to a section within the document.

• Managed Objects folder: If you are creating MDX cube training metrics, the training metrics are created in the Managed Objects folder of the MicroStrategy project.

• Automatically create on report execution: Select this check box to enable the automatic creation of predictive metrics feature to create predictive metrics automatically each time a report that contains the training metric is executed. This check box is selected by default.

  If you are creating a derived training metric, the Automatically create on report execution check box is selected by default and cannot be cleared. This is to ensure that predictive metrics can be created for the derived training metric, since derived training metrics are only available as part of a report or document.

  If a predictive metric exists with the same name and in the same location as specified in the training metric, it is overwritten automatically. Do not select this option if you do not want the predictive metric to be overwritten.
  

  If this check box is cleared, predictive metrics can be generated through the Create Predictive Metric(s) dialog box, available from the Data menu on the Report Editor. This menu also provides an option to save the generated PMML model to disk.

• Score Training Records: Select this check box to score the records that are used to train the predictive metrics. While the score can provide additional information about the predictive model, creating these scores requires additional resources. Clearing this check box can reduce the amount of resources required by excluding the creation of these scores for records.

  This check box is selected by default, except when performing Association rules analysis.

• Include extended statistical analysis with the model: Select this check box to include statistical analysis, in the model for the training metric, that is not directly used for scoring purposes. This check box is selected by default.

  If this check box is cleared, some statistical analysis that is not directly used for scoring purposes is excluded from the training metric's model. This can simplify the definition of the training metric and can provide improved performance in some scenarios, especially for large Tree regression models. For a list of the statistical analysis that is excluded for each model type if you clear this check box, see Excluding extended statistical analysis below.

• Number of verification records to include with the model: Use this setting to specify the number of model verification records that should be included within the generated PMML model.

• Predictive metric(s) to generate: Use the check boxes in the Predictor Type column to select the desired outputs. An individual predictive metric will be created for each selected output.

  A training metric can be set up to produce a variety of outputs, based on the selected type of analysis. At least one type of output must be selected in order for the training metric to be created. The following table lists the types of outputs supported by each type of analysis:
   

  Allowable Inputs Based on Analysis	Linear or Exponential	Logistic	Cluster	Decision Tree	Time Series	Association Rules
  Predicted value	Yes	Yes	Yes	Yes	Yes
  Probability	Yes	Yes		Yes
  Cluster affinity			Yes
  Entity ID			Yes	Yes
  Rule						Yes
  Antecedent						Yes
  Consequent						Yes
  Support						Yes
  Confidence						Yes
  Lift						Yes
  Leverage						Yes
  Affinity						Yes
  Rule ID						Yes

  Each predictive metric should include an aggregation function to ensure proper results when drilling and aggregating from a MicroStrategy report. The aggregation function is already set to the recommended function based on the type of values produced by the predictive model. However, you can change the aggregation functions for each output type by using the drop down boxes.

  If non-leaf level metrics from different dimensions are used as Independent metrics and Segmentation metrics, you may need to select None as the aggregation function to avoid unnecessary calculations and performance impacts. For example, if Quarter (from the time dimension) is used as an independent variable and Region (from the geography dimension) is used as a Segmentation metric, this can cause multiple calculations for the regions of each quarter.

• Rules (association rules analysis only): This button opens the Rules To Return dialog box. It provides options for setting additional criteria which will be used to determine what outputs the generated model will support. For more information, see Association rules analysis.

Once you have made all the required selections to determine the output of the training metric, do one of the following:

• Click Next to view a summary of the training metric you will create.

• Click Finish to bypass the Summary step and create the training metric. A dialog box prompts you for the location within your MicroStrategy project to save the generated training metric.

Excluding extended statistical analysis

The term Excluded in the table below indicates that a statistical analysis that is otherwise included for the model is excluded if you clear the Include extended statistical analysis with the model check box:

The absence of the term Excluded in the table below does not indicate that the statistical analysis is included for the model. Only relevant statistical analysis is created for each model. This table is intended to only list which models are excluded for certain models when clearing the Include extended statistical analysis with the model check box.

Statistical Analysis	Linear and Exponential Regression	Logistic Regression	Decision Tree Regression	Cluster	Association Rules	Time Series
R2				Excluded		Excluded
Adjusted R2	Excluded
p-Value (HighR2byChance)	Excluded
Likelihood Ratio Test - Intercept-only Model		Excluded
Likelihood Ratio Test - Final Model		Excluded
Likelihood Ratio Test - p-Value		Excluded
Likelihood Ratio Test - Degrees of freedom		Excluded
Degrees of freedom (df)	Excluded
RMSE (SEy)	Excluded					Excluded
Sum of Squares (SSE)	Excluded					Excluded
Sum of Squares Regression (SSR)	Excluded					Excluded
Sum of Squares Total (SST)						Excluded
Error Sum of Squares (ESS)				Excluded
Total Sum of Squares (TSS)				Excluded
F-statistic (F)	Excluded
Lift Chart	Excluded	Excluded	Excluded			Excluded
Confusion Matrix		Excluded	Excluded
Cluster Model Quality				Excluded
Number of Records	Excluded
Mean	Excluded	Excluded				Excluded
Standard Deviation	Excluded	Excluded				Excluded
Median	Excluded	Excluded				Excluded
Mode	Excluded	Excluded	Excluded	Excluded
Min	Excluded	Excluded				Excluded
Max	Excluded	Excluded				Excluded
Inter-quartile range	Excluded					Excluded
Total	Excluded	Excluded	Excluded	Excluded	Excluded	Excluded
Missing	Excluded	Excluded	Excluded	Excluded	Excluded	Excluded
Cardinality	Excluded	Excluded	Excluded		Excluded	Excluded
Analysis of Variance (ANOVA)	Excluded
Importance	Excluded	Excluded
Standard Error	Excluded	Excluded
z-Score		Excluded
p-Value(T-Dist)-Final	Excluded
p-Value(T-Dist)-Initial	Excluded
p-Value(ChiSquare)		Excluded
t-Value	Excluded
Correlation Matrix	Excluded	Excluded	Excluded	Excluded