System Architecture#

The high level API is designed around a few key abstractions and their corresponding types and interfaces. In addition, Rubix ML employs various mid and low level subsystems that power many of the learners. This layered architecture allows for power and flexibility while keeping the public interface simple and straighforward.

General Architecture#

From the perspective of data flowing in and out of a machine learning system, there are a number of components that the user may interact with. These include Dataset objects, Transformers, Estimators, and Meta-estimators. Starting from the top, the illustration below shows the path of data from input features to prediction within Rubix ML.

Rubix ML System Architecture


Under the hood, Rubix ML utilizes a number of modular subsystems that are highly optimized for their purpose such as the graph, neural net, SVM, and tensor computing subsystems. Some mid and low level subsystems run as optional PHP extensions.