search

AlineTalhouk / splendid

Supervised Learning Ensemble for Diagnostic Identification
https://alinetalhouk.github.io/splendid/
Other
1 stars 0 forks source link

What to do with list of best algorithms #1

Closed dchiu911 closed 7 years ago

dchiu911 commented 7 years ago

Hi @AlineTalhouk Using RankAggreg, I am able to get a list of the top performing algorithm for each bootstrapped resample. For example, if n.boot = 3, then we might get something like "svm", "svm", "rf". The paper describes a majority vote approach for the ensemble classifier in a new sample X (independent data). So I'm wondering what we want to output for the splendid() function.

Currently, I implemented the majority voting for ensemble class prediction on the entire, original dataset. But certainly there is overfitting there.

AlineTalhouk commented 7 years ago

I think we discussed this yesterday and we decided to output the models as list