Getting Started

Experimenting with MLeap is easy to do either through an interactiveScala console, Python console or a notebook.

A typical MLeap workflow consists of 3 parts:

1. Training: Write your ML Pipelines the same way you do today
2. Serialization: Serialize all of the data processing (ml pipeline) and the algorithms to Bundle.ML

As you will see from the basic usage section, most often all you have to do is import some MLeap libraries and that is it (except for scikit-learn).

Serialization

Once you have your pipeline trained, MLeap provides functionality to serialize the entire ML/Data Pipeline and your trained algorithm (linear models, tree-based models, neural networks) to Bundle.ML.Serialization generates something called a which is a physical representation of your pipeline and algorithm that you can deploy, share, view all of the pieces of the pipeline.

Did we mention that MLeap Runtime is extremely fast? We have recorded benchmarks of micro-second execution on LeapFrames and sub-5ms response times when part of a RESTful API service.

Note: As of right now, MLeap runtime is only provided as a Java/Scala library, but we do plan to add python bindings in the future.

Components

MLeap Spark

MLeap Bundle