H2O Scorer mechanics

Hello!

I have recently started to use H2O nodes to test out GBM learning. I’m trying to do classification on binomial class with one feature and around 800 values of that feature (processed through one-hot encoding).

While trying to use the Scorer node, it does not work when I pick prediction column with positive class. KNIME gives me error ‘Problem during statistic extraction. Maybe wrong column selected?’. However, when I use negative class, the score calculates correctly.

Can I use negative class score as a guidance for my model anyway, as it is binomial class prediction? Is this a normal behavior of Scorer node?

Hi @Kirky,

could you share your workflow with us? This sounds like a bug on our side.

Cheers
Simon

Hi @SimonS,

Sure, here it is along with the data:

H2O Test.knwf (33.0 KB)
Data Sample.xlsx (942.8 KB)

Thanks a lot! This seems to be a bug we need to fix.

I think in some sense this is possible, yes, but you must be careful with some scores which must be interpreted differently.