I need to implement an RF created by R on Hadoop. I know a way is to recreate the random forest with the same training data using Mahout. But is it possible to convert the RF created by R directly into a format that Mahout can use, for scoring?
One approach would be to use the Rattle package in R to export your Random Forest model as PMML. Then use Cascading to score with that PMML model at scale on Apache Hadoop.
Examples for just this are shown at:
Code is already available in the pre-defined app. Just run that JAR shown in the example.
A presentation about this is showing at slide 48-ff in http://www.slideshare.net/pacoid/the-workflow-abstration