Bug? Training SVM with Nominal Label. Issue with Nominal to Binominal Operator
UnicornHi everybody,
I am checking the handout for the course "Basics Part 2", and I am running into an error when training the SVM model.
The label is called Churn and it has been generated with the Generate Attributes operator:
if(missing(ChurnDate), "loyal", "churn")
This results in a Nominal attribute, later set to role "Label".
When I try to train SVM it warns me that it only works with binominal labels. Then I decided to add a Nominal to Binominal in the ETL process. However after performing this transformation, the Set Role Operator throws the following error:
However the attribute IS present, as shown by putting a breakpoint after Nom to Bin:
Furthermore if I Execute the ETL process as the first steps in the training process, I get an error that label information is missing (but surprisingly I am able to train the model anyway):
But the strangest thing is that if I store the data set at the end of the ETL step, and then recall it (called svm temp in the last picture) everything runs ok. I think there is some kind of problem with the Nominal to Binominal Operator, what do you think about it?
Best regards,
Sebastian
Answers
Dear Sebastian,
a general thing is, that the difference between nominal and binominal is rather a meta-data thing than a real problem. The meta data tells the SVM that this is nominal. Once the data "arrives" RM can check and realize, that the data is nominal, but only has two classes. So it can run. This is why the SVM produces no error.
The thing with set role is a bit odd. can you sent me the process or the full repository so I can have a look?
Which training did you attend? The last basics 1 in Dortmund?
Best,
Martin
Dortmund, Germany