The KNIME hub contains an example called “02_Document_Classification”. In this example the training of a text classifier is illustrated which is followed by a deployment pipeline. When applying this to my own dataset, I struggled with getting the dictionary structure that is loaded in with the Table Reader in the deployment pipeline. Where does this dictionary come from? Is this something that I have to extract from the training workflow and if yes, how is this done?
The structure of the dictionary is indeed created by the Document Vector node in the training branch.
I didn’t write this workflow, but I suspect it may have been written before the Document Vector Applier node was in wide use. That node is an easy way to sidestep this issue, using the model output ports.