Options

ROC Comparison for polynominal label

PosaPosa Member Posts: 1 Contributor I
edited October 2019 in Help
Hi all.

I wanted to create the ROC comparison for several models using the Iris data set which has 1 nominal label attribute with 3 values. I thought that just by putting the ROCComparison operator after loading the Iris data from the repository will do the work, i.e. that it will provide actually 3 comparisons, one for each nominal value. Instead it just complained about too many values of the label.

So, I thought that I will first transform the one label attribute into 3 binominal atributes, and then will use the Loop Labels operator to create one ROC comparison for each binominal label. But:

When I transformed the label attribute to 3 binominal ones, the operator removed the "label" role from them. If I then tried to assign the "label" role to the 3 new attributes, the operator complained that only 1 label attribute can be present in the ExampleSet. But if only 1 label attribute is allowed, what is the purpose of the Loop Labels operator?

Or, is there any other, better way of creating the ROC comparison for each value of the label attribute?

Thanks a lot, Petr
Tagged:
Sign In or Register to comment.