Simulation tab, Simulator sub-tab

The Predictive Model Viewer includes a simulator that allows the behavior of the model to be explored interactively. This simulation does not operate on actual Customer data but rather allows the user to explore the possible ranges of variables and the corresponding model response.

For details on the Simulator tab for Association Rules models, see Simulation tab for Association Rules.

Simulation Graph

The simulator scores models by passing in a value for each variable and graphing the result. This means that input variables are separated into a range of discrete values that can be used in scoring, one at a time. There is a limit to the number of discrete values so that simulations can run to completion in a reasonable amount of time. The Predictive Model Viewer will inform the user when this limit is exceeded.

This simulator displays results in a multi-dimensional graph. The results are color-coded which makes it easier to understand how the behavior of the model changes in response to changing inputs. The results are also available in a grid format on the Performance, Model Response, and Static Variables tabs.

The user has control over which variables are placed on each axis of the graph using the radio button associated with each variable. The output of the model, the predictor, can only be placed on the Y-Axis.

To the left of the simulation graph is a list of variables that comprise the model.

The predictor (that is, the variable being predicted) is always at the top and contains a legend that shows the range of output values possible from the model, and the corresponding colors used. In some cases, this range can be determined from information in the PMML and, if so, there will be values in this legend prior to performing a simulation. For example, the number of clusters in a cluster model can always be determined from the PMML, and therefore this legend will always be populated. There are other cases, however, where this range is not known prior to executing the model. This happens when the PMML does not include any information about the range of outputs possible from the model. In these cases, the Predictor legend will only appear once the simulation is executed and the range of results can be determined. This also means that this range can change with each simulation run, as the values on inputs change. Note that the predictor can only be placed on the Y-Axis.

The remaining variables appear below the predictor. Users can specify whether the variable is on the X-Axis, Y-Axis, or the Slicer. The Slicer specifies which "slice" of the solution space is visualized. This slice can be changed by moving the slider.

There must always be at least one variable on the X-Axis and at least one variable on the Y-Axis.

For each variable, clicking the Edit button allows the user the option of editing the range of values and the data type. If the PMML defines the range of values for the variable, then the appropriate slider will have this range. If not, the viewer will default to a basic range, such as 1 to 10, incrementing by 1. Each variable has an associated Value List which contains the values used in the model. To add a single value, click theAddbutton. To remove a value, select that value and click theDeletebutton. To remove all values, click theClearbutton.

For numeric data, you can specify a minimum and maximum value, along with an increment amount. This range can be added to the Value List, or replace the Value List by clicking the corresponding button.

These changes are not saved in metadata; therefore, closing and re-opening the viewer will restore the original settings.

Simulation Execution

The simulator has the following modes:

• In Manual mode, the user clicks the Simulate button to execute a simulation. The user can change the variables on each axis or they can move the sliders and click Simulate to see a different "slice" of the result space.

• In Automatic mode, any time a slider or axis variable is changed, the simulation automatically executes.