--- jupytext: text_representation: extension: .md format_name: myst kernelspec: display_name: Python 3 language: python name: python3 --- # Installation `aicad` is a Rust crate, and `pyaicad` is the Python package built from it. Install the one you need. ## Python ``` pip install pyaicad ``` `pyaicad` comes in two builds of the same package. The **slim build** has the modelling, the compiler and the diagrams, and the local search with belief propagation (`BeliefPropagationSearch`). It has no heavy dependency. The **full build** adds training and the neural searches (`NeuralPriorSearch`, `NeuralPriorBPSearch`, `pyaicad.learning`). It embeds the deep-learning framework [burn](https://burn.dev), which takes a long time to compile. `pyaicad.HAS_LEARNING` tells which one is installed. Asking the slim build for something that needs the full one raises `FeatureUnavailableError`, with a message saying so. ```{code-cell} python import pyaicad as ac print(ac.__version__, "- learning included:", ac.HAS_LEARNING) ``` ## Rust ``` cargo add aicad # the compiler, belief propagation, local search cargo add aicad --features learning # and the neural networks, their training and the neural searches ``` The `learning` feature pulls in [burn](https://burn.dev), which takes a long time to compile, and is off by default, as for the Python wheel. The code that needs it is gated with `#[cfg(feature = "learning")]`, and the Rust guide marks the chapter that needs it. ## Building from source From a checkout, with [maturin](https://www.maturin.rs): ``` maturin develop # slim build maturin develop --features python,learning # full build ``` A `--features` given on the command line replaces the list in `pyproject.toml`, so `python` has to be named again. To run the Python tests, `pip install -r requirements.txt`; to build this documentation, `pip install -r docs/requirements.txt` and `sphinx-build -W docs docs/_build` (it needs the full build, since the guide runs the training).