I was wondering whether it is possible to make available an Ordinal Regression Learner Node and corresponding Predictor Node. Right now we have only Logistic Regression Learner Node and Predictor.
Appreciate an early reply on this, since the requirement is rather urgent.
However, on your suggestion, I tried to plug in the linear regression node and found that it doesn't recognize the string fileds at all. I am not sure, if I convert the ordinal values in to numbers, this issue will get resolved, as also its implications from the regression perspective.
You could use the "Category to Number" node and transform your ordinal values into integers. You have to make sure they are in the correct order before feedings them into the node. Then you can use Linear Regression on the newly created numeric column. It's not a real ordinal regression, but we don't have appropriate node yet. Then you need to round the predicted regression values to integers (using "Double to Int") and finally join this table with the original table that contains both the original categories (ordinal value) and their artificial numerical values.
I read somewhere that the linear regression and the ordinal regression are not exactly the same.
The range of nodes for clustering and classifying from Weka will bear testimony for the same. I was wondering if you could give some pointers to extracting the regression and other statistics from some of the relevant Weka nodes. Alternately, I could explore with them directly too.
Isn't ordinal regession trainer and predictor not in your roadmap?
Thanks for your clarifications, including that ordinal regression learner is not in KNIME roadmap.
It is only logical that if there is a logistic regression learner, there is also a ordinal regression learner. I guess my need is not unique. I did find ordinal regression is commonly used - for e.g. medical field. In the field of management, ordinal dependent variables are pretty common.