Running a quick example¶

This page explains how to install RAITAP itself and how to get it running quickly, with a simple demo example. If you want to see how to fully configure your own assessment, skip to the Using RAITAP in your project and Creating & running your own configuration pages.

It is recommended to use uv, but pip will also work.

1. Install RAITAP¶

First, you need to install the RAITAP package itself.

uv add raitap

pip

pip install raitap

Note

RAITAP supports Python 3.11–3.13. Python 3.14 is not yet supported (Hydra 1.3.2 limitation). Some underlying libs require older versions (e.g. Marabou < 3.12). RAITAP will handle the interpreter choice for you.

2. Run the demo example¶

RAITAP ships with a self-contained demo.yaml you can run with a single flag. It uses a tiny bundled dataset and CPU execution, so it works out of the box on any machine.

uv run raitap --demo

pip

raitap --demo

RAITAP does not ship with all the underlying dependencies by default, to avoid massive bloat. The required dependencies are automatically inferred and installed by default.

In some specific setups, you might need to take action for the automatic install to occur:

On uv, RAITAP may ask for --allow-project-edit (or -y).
On pip without a venv, RAITAP may ask for --exec-global.

3. Inspect the output¶

After the run is complete, the outputs directory can be found in the directory you ran RAITAP from.

It will contain the run's metadata, the transparency assessment (attributions and visualisations), the robustness assessment (adversarial examples, per-sample verdicts, and an image-pair visualisation), and a PDF report under reports/. Refer to the Understanding outputs page for more details.