mleap

Why we included this project

MLeap tackles a familiar pain: a trained Spark or Scikit-learn pipeline that works well in a notebook but is hard to serve behind a real API. It serializes those pipelines into a portable Bundle format, and a lightweight Scala runtime executes the result on the JVM with no Spark, sklearn, numpy, or pandas in sight. You train once, export, and then either score in a realtime service or load the same bundle back into Spark for batch runs. Because the format spans Spark, PySpark, Scikit-learn, and TensorFlow, teams can mix transformers from different frameworks in one pipeline instead of committing to a single stack or cloud vendor. If your serving layer is JVM-based, this gives you one common way to run research pipelines in production without reimplementing them.