How to export sparklyr (Spark ML) models to PMML?

Question

I know that Spark ML pipelines can be exported to PMML using the JPMML-SparkML library. I am just struggling to find out how I could do it from R using sparklyr.

I am aware of open github issue, where two ideas were raised:

using Scala API, something like:

model <- ml_kmeans(<...>)    
sparkapi::invoke(model$.model, "toPMML", "./myModelPMML.xml")

leverage https://github.com/jpmml/jpmml-converter and the https://github.com/jpmml/jpmml-sparkml

However I could not find any follow ups on that tips. Has anyone tried any of these solutions?

Here is the related github issue to this question.

You'll need to wait for the answer of sparklyr on github I guess since you asked the question there an hour ago. I don't believe it's possible for the time being. — eliasah, Jun 22 '17 at 14:45
Yes, I posted a question on github as well. I am just checking here if anyone tried any workarounds. — michalrudko, Jun 23 '17 at 11:29
I personally tried to find one. There is some limitations with sparklyr itself. — eliasah, Jun 23 '17 at 11:30

user1808924 · Answer 1 · 2018-07-06T08:50:50.670

0

You can use the sparklyr2pmml package to export Sparklyr/Apache Spark ML pipelines to PMML.

edited Jul 06 '18 at 08:50

answered Jul 05 '18 at 17:50

user1808924

4,563
2
17
20

How to export sparklyr (Spark ML) models to PMML?

1 Answers1

Linked