Explain Model Action Set
Frequency and Weight Variables
The freq parameter is an optional parameter that names a numeric variable that provides the frequency of each observation in the reference data set. If you do not specify a frequency variable, then the linearExplainer action assumes that the frequency of all observations is 1. The linearExplainer action expects nonnegative integer frequency values. If the frequency of an observation is not an integer, the action truncates the frequency value to the largest integer that is smaller than or equal to the value. The action treats negative and missing frequencies as 0.
The weight parameter is an optional parameter that names a numeric variable that provides the weight of each observation in the reference data set. If you do not specify a weight variable, then the linearExplainer action assumes that the weight of all observations is 1. The linearExplainer action expects nonnegative weight values. The action treats negative and missing weights as 0.
The linearExplainer action uses the frequency and weight variables differently, depending on the data generation method that you specify in the method subparameter of the dataGeneration parameter:
If you specify NONE, the action uses both the frequency and weight variables in the linear regression step. The action uses the frequency variable directly as the linear regression frequency variable. The action generates the linear regression weight by multiplying the weight variable by the calculated kernel weight that you specify in the
distanceparameter.If you specify QUERYCENTERED, the action uses both the frequency and weight variables to calculate the variable mean and variance for interval variables. For nominal variables, only the frequency variable is used in the sampling process.
If you specify REFERENCEDISTRIBUTION, the frequency variable is used in the sampling process for both the interval and nominal variables.
The linearExplainer action does not use the frequency or weight variable from the query data even if the variables exist. The query data are always treated as if they have a frequency and weight of 1.