I would like to know if it is possible to obtain class probabilities as a prediction in Apache Spark for the Spark MLlib models. The Spark Predictor (MLlib) produces a prediction but not class probabilities. Also, the Spark MLlib to PMML node does not support some Spark MLlib models.
unfortunately for the “old” MLlib nodes it is not possible to obtain class probabilites. It is possible for the spark.ml based nodes however:
There are plans to revise more of the Spark learner nodes to use spark.ml, but currently no concrete date has been set for this.
Not a solution for this problem but you could use models from H2O.ai also on a Big Data cluster with Sparkling Water, like this: