I'm going to apologize in advance incase I've missed something that already exists in the platform.
For the implemented learners (and Weka learners) which do not have a model output / separate predict method how does one score new data without having to retrain the model?
unfortunately, there is not a way to classify the data without retraining if there are no model ports available.
I'm just aware of the KNN node that does not have a model outport (because it does not make much sense). We are working on providing the weka nodes with a model outport and a generic weka predictor.
I hope this solves your problems in the next version.
Will the Weka models be stored in the same format Weka uses? I'm wondering if models previously built with weka itself or another weka interface could be loaded and used to score a dataset in Knime?
the new weka nodes store the classifier at the model port (among with additional meta information) in the same format as Weka is using, which is a serialized object. It will be included in the next release of Knime.
The suggestion to load previously build models from weka into Knime is very interesting, thank you! I will take a look into that issue and let you know if it is possible.