Basics

This section of the documentation introduces the main concepts of UnionML and aims to show you the core pieces that together make up a UnionML app.

UnionML Apps

A UnionML app is composed of two objects: Dataset and Model. Together, they expose method decorator entrypoints that serve as the core building blocks of an end-to-end machine learning application. The following sections will show you how to:

• Initialize a UnionML App: quickly create a UnionML app with unionml init.

• Define a Dataset: specify where to read data from and how to split it into training and test sets, parse out features and targets, and iterate through batches of data.

• Bind a Model and Dataset: specify how to initialize, train, evaluate, and generate predictions from a model given a Dataset.

• Train and Predict Locally: perform training and prediction using a UnionML as a regular python script, then serve a FastAPI app for prediction.