Create Ensemble Model Process with existing Models?

Fred12Fred12 Member Posts: 344 Unicorn
edited November 2018 in Help

hi,

I would like to create an ensemble of models like with voting or stacking, but from existing models, that means, I already trained models from my dataset for k-NN, SVM and several Decision trees and saved them in the repository. Now, I would like to create an ensemble of those models and test them on new test data with a model voting scheme or stacking, or whatever other possibilities there exists.

 

The tutorial processes, show only processes where you are still training the models on train data... But is it possible with existing models, too? Some example process or explanation how it works would be nice.. :)

Answers

  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,503 RM Data Scientist

    did you try to retrieve them on the left hand side of stacking and connect them?

     

    ~Martin

    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • Fred12Fred12 Member Posts: 344 Unicorn

    yeah I did so,

    I connected 3 models, with level 1 GLM Model learner, but my stacked model results were not better, in fact slightly worse than each of my individual models... is that normal?

  • Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,761 Unicorn

    Not all the time the ensemble models will be better, but your results would give me pause and see if I made an error somewhere in the process.

  • Fred12Fred12 Member Posts: 344 Unicorn

    ok, one more question...

    is multiple stacking possible? I Mean level1 base learners, then several Level2 Learners , and on top of that a level3 learner?

  • Thomas_OttThomas_Ott RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,761 Unicorn

    I don't see why not! Since they're subprocesses you could embed new Stacking operators inside them. 

Sign In or Register to comment.