[SOLVED] Vote operator. How it works?

tobix10tobix10 Member Posts: 7 Contributor II
edited November 2018 in Help
Hello, I have a process where inside vote I split my data set into 5 parts with different attributes and each part goes to SVM classifier. These 5 output models are outputs of vote operator.
Everything is done in X-Validation and testing subprocess contains Apply Model -> Performance(typical approach).

I've read description of Vote Operator and it says that it uses majority vote to take prediction. My process is classification problem(only 2 classes: true, false).
I am wondering how vote exactly works.
1. Does it take for each record only predicted class from SVMs outputs and count how many times it was true and false, and take the largest one(ex. 4 SVM predicted true class for the record so vote also predicts true)?
or
2. It uses some kind of probability approach. I mean that SVM classifiers gives an information about how likely record is true and how likely it is false and vote learns on that predictions.

If first is correct how can I perform second approach ?

Answers

  • earmijoearmijo Member Posts: 270 Unicorn
    RapidMiner uses approach 1 (it makes the different algorithms vote and assigns the observation to the class that got more votes).

    If you want to average the probabilities you can use the Vote Operator available from Weka. With this one you have the choice (simple voting or averaging probabilities)
  • tobix10tobix10 Member Posts: 7 Contributor II
    Ok, thx for reply.
  • akunyerakunyer Member Posts: 3 Contributor I

    Hi , how to know which algo that the Vote method choose inside RM ? Thanks.

  • sgenzersgenzer Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,959 Community Manager
    Hello - this is a very old thread. Have you looked at the documentation for Vote (https://docs.rapidminer.com/latest/studio/operators/modeling/predictive/ensembles/vote.html) and the tutorial?

    Scott
  • akunyerakunyer Member Posts: 3 Contributor I

    Yes, I have read the documentation and also already implemented it. No issues. But the Vote techniques seems kind like a black box technique which only produced the output without knowing which one is having the maximum accuracy. Btw, I'm using the w-Vote which is from Weka.

     

    vote.png

     

    vote2.png

     

    vote3.png

     

    The accuracy result above is 86.63% which is good, but the result coming from which classifier? I'm set as "MAX" in the W-Vote which it should return the maximum accuracy among the selected classifier.

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

    Hi,

     

    nono. it takes the classifier with the highest confidence. Basically you have 5 (or k) models producing 5 (or k) confidences.

    Vote takes the 5 and merges them to one. usualy via an average. you used the maximum, which can also work.

     

    Best,

    Martin

    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • akunyerakunyer Member Posts: 3 Contributor I

    Yes, i know it takes the highest , but how to check which classifier it choose using RapidMiner ? I really need that for my research .

Sign In or Register to comment.