Best classification for SVM?

bdobbsbdobbs Member Posts: 2 Contributor I
edited September 2019 in Help
I am just learning RM, so forgive me if my questions are too "simple".

1) I want to train a two class SVM.  When generating my training data, I can label it as "1,0" or "true,false".  Does it matter?

2) I've read about AUC as a performaance measurement.  That sounds good.  How do I tell RM which class is true/false so that the AUC is calculated correctly?  In my tests, it makes the positive class "false".  That seems backward.  Since I'm just loading in a labeled CSV example set, where do I tell it which class is which?

Answers

  • bdobbsbdobbs Member Posts: 2 Contributor I
    I figured it out! 

    A bit of a "Long workaround", but it works cleanly.

    1) open my data file with CSVExampleSource
    2) Save my data file with save ARFF Example Source
    3) Edit the ARFF file and correct the order of the nominal items for my label attribute
    4) Run RapidMiner and everything works correctly!

    I guess there isn't a way to determine nominal class order from a CSV file.  By converting it to ARFF I was able to make it all work well.
  • IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder
    Hi,

    yes, that is option 1, another one is to use the .aml file of RapidMiner and change the order of the nominal value in this meta data file. And the third option is to simply change the order of lines in you csv file - the negative class has to be the first, the positive second. Of course the third option is only a lazy man's option  ;)

    In future releases, the performance operators for binominal measures will have a parameter for the positive class and there will be operators to perform those changes as well.

    Cheers,
    Ingo
Sign In or Register to comment.