PMML is a standard developed by the Data Mining Group (DMG). The standard enables inter-platform interoperability by letting you build on one system and then deploy to another system in production. The PMML standard has gained momentum and has been adopted by most vendors. At its core, the standard is based on an XML document with the following:

• Header with general information
• Dictionary describing field level definitions used by the third component (the model)
• Model structure and parameters

As of this writing, the Spark 2.0 Machine Library support for PMML exporting is currently limited to:

• Linear Regression
• Logistic Regression
• Ridge Regression
• Lasso
• SVM
• KMeans

You can export the model to the following file types in Spark:

• Local filesystem:

Model_a.toPMML("/xyz/model-name.xml")

• Distributed filesystem:

Model_a.toPMML(SparkContext, "/xyz/model-name")

• Output stream--acting as a pipe:

Model_a.toPMML(System.out)