Using the api from elasticsearch-prediction, this is the implementation that uses Spark as the backend to compute large scale models offline, and generates a plugin for elasticsearch for runtime evaluation of a trained scoring function.
This is highly experimental code as a proof of concept, so there are many many areas of improvements, and bugs. Use for fun only
Note that currently the only suppported spark models are linear, until the serialization and spark.ml API matures out of beta
Some references:
Code is provided under the Apache 2.0 license available at http://opensource.org/licenses/Apache-2.0, as well as in the LICENSE file. This is the same license used as Spark and elasticsearch-prediction.