Once you have identified your filtering outputs, you can filter and sort your inputs with regards to their influence on the filtering outputs. Input parameters are categorized either as major inputs (with a greater influence on the outputs) or minor inputs (with a smaller influence on the outputs).
To specify your filtering criteria, select the Parameters Correlation node in the Outline view and set properties under the Filtering Method section of the Properties view. The following properties are available:
Relevance Threshold: Determines how strictly inputs are filtered for inclusion in the major input category. To estimate the relevance of parameter relationships, a metric is computed for any input-output pair. If one of the metrics for a given input parameter exceeds the value set for Relevance Threshold, that parameter is categorized as a major input. Otherwise, it is categorized as a minor input.
The default value is 0.5, with possible values ranging from
0 to 1. A value of
1 applies the strictest filter, and a value
of 0 the most relaxed. For example, if there are
10 inputs and Relevance Threshold is set to
0, all 10 of the outputs are categorized as
major outputs. If Relevance Threshold is changed to
1, a number of the outputs are filtered out
and categorized instead as minor inputs.
For more information on how Relevance Threshold filters inputs, see Parameters Correlation Filtering Theory.
Correlation Filtering: This is based on Pearson's, Spearman's, and quadratic correlation. For each input-output pair, both correlation complexity and the number of samples are used to determine relevance. The greater the number of samples, the greater the probability that the correlation detected is valid. If the relevance is greater than the Relevance Threshold value, the input is categorized as a major input. This property is enabled by default.
R2 Contribution Filtering: Computes a reduced model and evaluates the R2 contribution of each input parameter on the filtering output. The R2 contribution is then used to calculate relevance. If the relevance is greater than the Relevance Threshold value, the input is categorized as a major input. This property is enabled by default.
Maximum Number of Major Inputs: Maximum number of input parameters that can be categorized as major inputs.
The maximum value is equal to the number of enabled input parameters. The minimum value is
1. The default is 20
if the number of input parameters is greater than or equal to 20. This
is the recommended maximum number of inputs for a response surface.
Once you have set your filtering criteria, click to filter the results. If design points have already been updated, the Update operation does not update design points again. It only updates the results according to the new filter. This allows you to change your filter settings and review multiple scenarios without having to run a new design point update each time.
Note: When you select only one filtering method (Correlation Filtering or R2 Contribution Filtering), the relevance threshold can change the list of major input parameters but not their best relevance values.
When you select both filtering methods, Correlation Filtering is applied first, which retains a list of major input parameters based on the Relevance Threshold value. The R2 Contribution Filtering method is then applied, based on the list of major input parameters retained by the Correlation Filtering method. This can cause the best relevance values displayed in the tables of major and minor input parameters to change slightly when you change the Relevance Threshold value with both filtering methods enabled.