Transforms and Pipelines

The central abstractions in PADL is the Transform. A Transform is a function with added functionality for the deep learning workflow.

Transforms can be combined to form Pipelines.

With Transforms you can:

• build complex workflows with an elegant formalism and with minimal boilerplate code

• preprocess data using multi-processing without creating custom DataLoaders

• stop worrying about adding batch dimensions, dis- or enabling gradients or sending data to the GPU

• work interactively in a notebook, easily inspect, slice and debug your transforms

• save your whole workflow in a transparent, flexible format that enables reproducibility

and much more.

Read the next section to learn how to create Transforms.