Flexible model composition in machine learning and its implementation in MLJ

TL;DR

This paper introduces a graph-based protocol called learning networks for flexible model composition in machine learning, implemented in the MLJ framework, overcoming limitations of existing platforms and encompassing methods like model stacking.

Contribution

It presents a novel graph-based protocol for model composition, with a concise syntax in MLJ, enabling flexible inclusion of various ensemble methods like stacking.

Findings

Learning networks overcome limitations of existing model composition methods.

The protocol is implemented in MLJ with a concise syntax.

Learning networks can include model stacking with out-of-sample predictions.

Abstract

A graph-based protocol called `learning networks' which combine assorted machine learning models into meta-models is described. Learning networks are shown to overcome several limitations of model composition as implemented in the dominant machine learning platforms. After illustrating the protocol in simple examples, a concise syntax for specifying a learning network, implemented in the MLJ framework, is presented. Using the syntax, it is shown that learning networks are are sufficiently flexible to include Wolpert's model stacking, with out-of-sample predictions for the base learners.