Is it possible to export a boosting ensemble model to PMML?

KNIME Analytics Platform
1

Hi guys,

I'm using the boosting learner and predictor meta nodes. I can correctly use them for create a model and make predictions for a certain data set. These meta nodes are very useful, anyway I'm not finding any way to export this boosting ensemble model to PMML. Is that possible?

In case of Tree Ensemble Learner node I can export the corresponding PMML ensemble model by using the Tree Ensemble Model Extract node, followed by Table to PMML Ensemble one. Anyway this solution is not compatible with the Boosting Learner meta node output (see attached workflow).

Alternatively if I connect the boosting learner meta node to a Table to PMML Ensemble node or to a Cell to PMML one (see attached workflow), it says me: No column in spec compatible to "PMMLValue".

Any suggestion is appreciated.

Thanks in advance,

Gio

2

Is there any solution to this post? I am performining the same procedure. I build a gradient boosting by using the boosting learner node and the boosting predictor bode. I don't know how to connect the boosting node to PMML node. 

Thank you.

3

Hi Ckuo,

Unfortunately I still didn't find a solution for this issue. I hope KNIME people could introduce this new feature in future releases.

4

Hi,

to convert the models coming out of the boosting meta node, you need the "Cell to Model" node. It can only deal with a single row, so to generate an ensemble from the models, you do the following:

"Boosting Learner" -> "Chunk Loop Start" (default settings) -> "Cell to Model" -> "PMML Ensemble Loop End"

The "PMML Ensemble Loop End" then outputs the ensemble.

Regards,

Alexander