01-18-2017 08:20 PM - edited 01-18-2017 08:22 PM
Given a process including cross validation involving n attributes, I would like to iterate the process each time with n-1 attributes (which becomes n combinations) and for each combination, with 10 different local seed values in the CV subprocess. In total that becomes n * 10 combinations.
The "loop attribute sets" and "loop parameteres" operators are out there to be used for this purpose, however, I need to write the results somewhere (let's say to a DB) for each combination alongside with the corresponding CV local seed value as well as the missing attribute (the excluded attribute in the n-1 set). Is there a way to "get" the current parameter / excluded attribute in those two looping operators, so I can accomplish this mission?
01-19-2017 03:58 AM
You could try the Backward Elimination operator. It does exactly what you want: starting with all attributes, taking off one after the other etc.
It won't set the name of the eliminated attribute by default, but there should be ways around that.
From the top of my head:
Main process: (your dataset) => Data to Weights => Remember (io object = AttributeWeights)
In the Backward Elimination => (current dataset) => Data to Weights => Recall full weights => Weights to Data on both => Left outer join => you get the attribute(s) missing from the main dataset.