Density Functions (simulation)

These settings allow you to customize output for probability density functions and cumulative distribution functions for continuous targets, as well as bar charts of predicted values for categorical targets.

Probability Density Function (PDF). The probability density function displays the distribution of target values. For continuous targets, it allows you to determine the probability that the target is within a given region. For categorical targets (targets with a measurement level of nominal or ordinal), a bar chart is generated that displays the percentage of cases that fall in each category of the target. Additional options for categorical targets of PMML models are available with the Category values to report setting described below.

For Two-Step cluster models and K-Means cluster models, a bar chart of cluster membership is produced.

Cumulative Distribution Function (CDF). The cumulative distribution function displays the probability that the value of the target is less than or equal to a specified value. It is only available for continuous targets.

Slider positions. You can specify the initial positions of the moveable reference lines on PDF and CDF charts. Values that are specified for the lower and upper lines refer to positions along the horizontal axis, not percentiles. You can remove the lower line by selecting -Infinity or you can remove the upper line by selecting Infinity. By default, the lines are positioned at the 5-th and 95-th percentiles. When multiple distribution functions are displayed on a single chart (because of multiple targets or results from sensitivity analysis iterations), the default refers to the distribution for the first iteration or first target.

Reference Lines (Continuous). You can request various vertical reference lines to be added to probability density functions and cumulative distribution functions for continuous targets.

• Sigmas. You can add reference lines at plus and minus a specified number of standard deviations from the mean of a target.

• Percentiles. You can add reference lines at one or two percentile values of the distribution of a target by entering values into the Bottom and Top text boxes. For example, a value of 95 in the Top text box represents the 95th percentile, which is the value below which 95% of the observations fall. Likewise, a value of 5 in the Bottom text box represents the 5th percentile, which is the value below which 5% of the observations fall.

• Custom reference lines. You can add reference lines at specified values of the target.

Overlay results from separate continuous targets. In the case of multiple continuous targets, this specifies whether distribution functions for all such targets are displayed on a single chart, with one chart for probability density functions and another for cumulative distribution functions. When this option is not selected, results for each target will be displayed on a separate chart.

Category values to report. For PMML models with categorical targets, the result of the model is a set of predicted probabilities, one for each category, that the target value falls in each category. The category with the highest probability is taken to be the predicted category and used in generating the bar chart described for the Probability Density Function setting above. Selecting Predicted category will generate the bar chart. Selecting Predicted probabilities will generate histograms of the distribution of predicted probabilities for each of the categories of the target.

Grouping for sensitivity analysis. Simulations that include sensitivity analysis generate an independent set of predicted target values for each iteration defined by the analysis (one iteration for each value of the input that is being varied). When iterations are present, the bar chart of the predicted category for a categorical target is displayed as a clustered bar chart that includes the results for all iterations. You can choose to group categories together or you can group iterations together.