Example using PMML output for batch prediction

Knime tech;

I have created a pretty good SVM model, and would now like to use it with new data. So far, my efforts to use the PMML output and new data to drive a post-training prediction process have produced exactly random results, unlike the 85% accuracy of the model.

Is there a source for workflow examples where the PMML product of a trained model can be used with new data for batch production of predictions? (This sounds pretty basic to me, and as a new Knime user, I apologize in advance if I missed some truly obvious answer.)

Thanks,

Kirk

 

Is there any pre-processing / normalisation going on prior to model building that doesn't happen when you use the model later?