Spark MLlib to PMML node

KNIME Extensions
,
1

I am trying convert Spark MLlib Model (Logistic Regression and Decision Tree) to PMML, using Spark MLlib to PMML node, but does not work.

>WARN Spark MLlib to PMML 0:432 Spark model Logistic Regression not supported

Please does anyone know how to resolve this issue?

Thanks

Marcus

2

Hello Marcus,

if you are using Spark 1.6 you could use the Spark Java Snippet Sink node to creat at least for a binary logistic regression model the pmml model using the Spark PMML model export function. You could write the file to HDFS and then retrieve it from their. However this function does not support the Decission Tree model: I guess the Spark Decission Tree model isn't compatible with the PMML spceification.

Bye

Tobias

3

Hi Tobias,

Thank you very much for your answer.

I am trying to run the logistic model (using MLlib) from data obtained in the Table to Spark node.
And I am getting some errors.

Please, do you have some example script to run this model from Spark Java Snippet node?

Thanks

Marcus

4

Hi Marcus,

attached you can find an example workflow that learns a regression model and writes it to HDFS.

Bye

Tobias

spark_pmml2hdfs.knwf
5

Hello Tobias,

Thank you vey much!

6

you are right…Spark Decission Tree model isn’t compatible with the PMML spceification.
So , is there any solution by now for converting it in PMML or can say for reading it?

7

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.