There are forecasting models, generated by different learner nodes inside other workflow. To interpret each model dedicated Predictor node is needed (e.g. Tree Ensemble Predictor/Weka Predictor).
- Is there any components (like “Universal Predictor”) or other solutions to interpret wide range of models (PMML Predictor don’t cover all possible variants)?
- When reading the model, how can I know it type to suggest dedicated Predictor node?
I think there is a good reason why there isn’t such a universal predictor. There are different types of predictions to make: e.g. numerical values, categorical values, probabilities or forecasts.
Why should this all be squeezed into one super-complex component, with all sorts of different output-types?
In the written model (a zipfile) is a file called object.file which is a zipfile as well. In there is the model.pmml. PMML is a version of XML (https://en.wikipedia.org/wiki/Predictive_Model_Markup_Language), so by scanning on specific tags one could determine the type of model contained in it.
An example of such a tag is:
<GeneralRegressionModel modelType="multinomialLogistic" functionName=... algorithmName="LogisticRegression" modelName="KNIME Logistic Regression" targetReferenceCategory=...>
But personally I would never go down that road. I would try to give those written models distinctive names which somehow show the type of model which is included. A model writer node is linked to a specific learner node, so it should be possible to create useful names.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.