

User Guide

AutoGOAL is a framework for the automatic generation and optimization of software pipelines. A pipeline is defined as a series of steps, which together form a program that performs some desired task. AutoGOAL was designed specifically for optimizing machine learning pipelines, a problem often called AutoML, but it can be used to optimize anything that can be defined as a set of steps with parameters.

AutoGOAL has been designed to suit a broad range of users with different skill levels, from beginners to experts. Likewise, the API suits different needs, from practical use cases requiring fast iteration and out-of-the-box solutions to more involved, research-oriented use cases that require customizing and tweaking many things. Whatever your case, the following guides should help you get started.

• Black-Box Optimization: A black-box optimizer that can be applied to any function.

• Predefined Pipelines: Pre-packaged with pipelines based on popular machine learning frameworks, that you can use in few lines of code to build highly optimized machine learning pipelines for a broad range of problems.

  • Using custom algorithms: Additionally, you can add your own implementations to the algorithm library.

• Class-based Pipelines: The class-based API allows you to turn any class hierarchy into an optimizable space. You define classes and annotate the constructor's parameters with attributes, and AutoGOAL automatically builds a grammar that generates all possible instances of your hierarchy.

• Graph-based Pipelines: The graph-based API allows you to explore spaces defined as graphs. You define a graph grammar as a set of graph rewriting rules, that take existing nodes and replace them for more complex patterns. AutoGOAL then transforms into an evaluatable object, e.g., a neural network.

• Functional Pipelines: If none of the previous suits you, the functional API allows you to magically turn any Python code that solves some task into an optimizable pipeline. You write a regular method and introduce AutoGOAL parameters in the code flow, which will be later automatically optimized to produce the optimal output.

Don't forget to also look at the examples for more down-to-earth specific use cases.