As far as I Understand it is a iterative search using Decision trees to select "good" attributes as judged by the Performance Operator. Is Loop Feature Subset therefor the best mothod to build such a workflow? Is there anything I have be take care of, for example setting of the minumum information gain etc.?