AUC Unknown

emtee
emtee New Altair Community Member
edited November 2024 in Community Q&A

Hi!

 

I'm doing predictive modeling using logistic regression with batch x validation. Everything is working fine I guess except for the AUC value. It shows Unknown as shown in the attached image. I also attached a screenshot of my process for your reference. Thank you!

Answers

  • MartinLiebig
    MartinLiebig
    Altair Employee

    Hi Emtee,

     

    kind of funny. I yesterday had an internal talk about Batch X-Val :).


    Are you sure that each fold of X-Val has both classes in it? Otherwise the AUC gets NaN, which might explain it. To check this you can either use breakpoints or use a store with something like ../data/%{a} as a path. %{a} will be replaced with the executions count of this operator (1 in the first iteration, 2 in the second etc). You have then k results in your repo and can check it.

     

    ~Martin

  • ozgeozyazar
    ozgeozyazar New Altair Community Member
    Hi @mschmitz !

    I also has this kind of output and I do not understand your solution advice. Could you please add more details ?

    By the way, I am doing classification with decision tree with AdaBoost and using cross validation.

    Best Regards,

    Özge
  • Telcontar120
    Telcontar120 New Altair Community Member
    If you have a small sample or a very imbalanced class, you could end up with a sample that does not contain both classes, in which case the AUC is not defined.  @mschmitz is saying you should check your samples and ensure that is not the case.   You can usually prevent this if you are using automatic sampling but if you are using other other sampling options it does not control for this.

  • ozgeozyazar
    ozgeozyazar New Altair Community Member
    Hi @Telcontar120

    thanks for explanation ☺️