Question on validation and AFE (Automatic Feature Engineering)

pblack476

Newbie here.

When using AFE I have to choose an operator to go inside the validation I pick in the subprocess. In my case that is Cross-Validation. But when I choose, let's say, a Decision Tree as my learner to cross validate and I get an optimal Feature Set from that. Does that Feature Set translate well if I wanted to apply Deep Learning or SVM or any other learner? Or the results I get from that Feature Engineering are only applicable to a Decision Tree in that case?

IngoRM

Hi,

sometimes features can be transferred but there is no guarantee for this.  The whole point of a so-called "wrapper approach" for feature selection / generation is to create the optimal feature set for that particular learner.  While this typically delivers better results, it also takes longer for many learners because they have larger runtimes and you need to experiment with the different options.  As always in life there is a tradeoff between resources spent and accuracy achieved...

Hope this helps,
Ingo