PMML Predictor new feature request

[gcincilla]

Dear KNIMERs,

I love the support that KNIME gives to handle predictive models in PMML format, anyway there are some things that can be improved. One of them is the options that the PMML Predictor node offers. This node is intended to be a general node to be used with any PMML file independently from the ML method used to build the model and that would be great. Nevertheless there are options in the specific PMML-compatible predictors nodes (e.g. PMML Gradient Boosted Trees Predictor , Naive Bayes Predictor, Logistic Regression Predictor, Rprop MLP Predictor, etc.) that are not present in the PMML Predictor node. More specifically I’m talking about the options of:

1. “Append columns with predicted probability” (giving the prediction probability for each class)
2. “Append overall prediction confidence” (giving the prediction confidence for the most probable class) (this option is present for exaple in the PMML Gradient Boosted Trees Predictor)

I think that including such options in the general PMML Predictor node would avoid to use specific predictor nodes. Would that be feasible?

Thanks in advance

Gio

[ipazin]

Hi there @gcincilla,

only configuration options that are common for all models can be added. Additionally have you checked JPMML Classifier node? It has option 1 available.

Br,
Ivan