Data Science Pilot Action Set

The generateShadowFeatures action performs a scalable random permutation of input features to generate shadow features that can be used to select all relevant features. The action generates these random permutations of input features without doing any sorting. Specifically, it executes a single pass through the data to profile the input variables by using an empirical cumulative distribution for continuous variables and an empirical frequency distribution for nominal variables. The action uses these profiles to do inverse sampling to generate the sample values. You can use the nProbes parameter to specify the number of shadow features per variable for the action to generate. The copyVars parameter copies variables from the input data table to the CAS output table.

The result of running the generateShadowFeatures action is a CAS output table that contains the generated shadow features, along with the input variables that are specified in the copyVars parameter.