Hands-On: The Future of Auto-Sklearn: A Modular AutoML Framework?

AutoML-Toolkit (AMLTK) is a new framework, designed to facilitate AutoML. It provides an extensive toolbox of components for building AutoML systems, including a declarative syntax to define pipelines and their search spaces, a suite of utilities and a common interface to HPO optimizers, and with the same code, scale from a simple notebook to a whole compute cluster. The AutoML-Toolkit was designed based on the experience of maintaining Auto-sklearn and Auto-PyTorch and studying the design of other AutoML systems.

The first part of the tutorial will be a short talk covering the motivation and core components of AutoML-Toolkit. Following this will be a hands-on-session where you will use AMLTK to:

• Define your own declarative parametrized scikit-learn pipeline

• Optimize hyperparameters of this pipeline

• Use AMLTK to make your mini-AutoML system extensible and scalable

This will allow you to build your own AutoML system from scratch with minimal overhead.