Polynomial to nominal?

Fred12

hi,

in the Logistic Regression, Sonar dataset has nominal data type as label... however when I import data from excel, I cannot choose data type nominal... however, logistic regression does not allow polynomial dataset as its not supported.. so what to choose then?

I find that really confusing.. in RM 5, you could choose nominal data type, is this only available in commercial releases or what happened?

is there a workaround for that?

Thomas_Ott

Yeah, there might be a problem in the meta data propogation, that happens sometimes but it should run.

Thomas_Ott

No, the functionality was not taken away in the free editions.

You can still change the data type in the import wizard by clicking on the header, or you can manually set it in the Read Excel operator.

Fred12

ok maybe I'm missing out something, but I still can't see an option to choose nominal data type (and some other types in previous versions, like numeric..):

no option for nominal...

Thomas_Ott

Ah maybe somewhere along the line from v5.3 the term nominal was changed to polynominal. There are two types of nominal attributes: binominal (true/false) or polynonminal (more than two nominal values). Logistic Regression will need you to change your column to binominal.

Fred12

ok and how is this best done if I have more than 2 class labels?

Thomas_Ott

Use the Nominal to Binomal operator.

Fred12

ok thanks, I will do that, however, there seems to be a difference between polynomial and nominal, when I am using nominal dataset to train my SVM model, and later testing a polynomial test dataset on it (which should be nominal and therefore, no difference), the Apply Performance operator gives a warning that my data should also be nominal instead of polynomial...