Pipelines

Machine learning pipelines are series of transformers that execute on a data frame. They allow us to combine our feature transformations together with our actual predictive algorithms. Pipelines can be as simple as a single transformer or quite complex, involving hundreds of feature transformers and multiple predictive algorithms.

Simple Pipeline Example

The diagram below shows a very simple pipeline that can be serialized to a bundle and then scored using MLeap Runtime. The ideas is that MLeap enables serialization and execution of transformers that operate on continuous and categorical features. A more complicated version of this pipeline may include dimension reduction transformers like PCA and feature selection tranformers like the Chi-Squared selector.

Advanced Pipelines

To see more advanced pipelines, please take a look at our MLeap demo notebooks.