Contributing to the data module

This page describes how to add new built-in dataset configs to RAITAP.

Overview

The data module provides built-in sample datasets that can be referenced by name in data.source. These configs live in src/raitap/configs/data/ and define metadata and default loading behavior for common datasets.

Adding a built-in dataset

To add a new built-in dataset config:

1. Create the config file

   Add a YAML config under src/raitap/configs/data/:

   # src/raitap/configs/data/cifar10.yaml
   name: cifar10
   description: CIFAR-10 — 60k 32×32 colour images in 10 classes
   source: cifar10
   

2. Add loading support (optional)

   If the dataset requires custom loading logic beyond what load_tensor_from_source() provides, add it to src/raitap/data/data.py or samples.py.

3. Use it

   uv run raitap data=cifar10
   

4. Update documentation

   Add the new dataset name to the list in docs/modules/data/own-vs-built-in.md.