We have a PMML model that we have developed in KNIME that we want to deploy elsewhere using a JPMML predictor. We are still in the debugging stage as we have noticed that the predictions probabilities made directly on the model in KNIME are very different from those we get on exactly the same data run outside of KNIME.
We are using the PMML model in conjunction with a 'SelectAll' multiple method in order to examine what's really happening as we can see the results of each tree within the model. Is there any way to do this within KNIME so we could output the probabilities of each tree in our test set?
Any thoughts on why we might also be getting different results with the JPMML predictors vs the in-built KNIME one would also be appreciated.
Thanks!