AI-eXtended Design (AIXD)

Introduction

In the current repository we collect the code for the general methodology for AI-augmented generative design. This methodology allows to invert the standard paradigm of parametric modelling, where the designer needs to tweak and tune the input parameters, iteratively or through trial and error, for achieving some desired performance values.

Instead, this method allows to, by just specifying the requirements' values, obtain a range of designs that closely approximate those. Besides, the present methodology allows the user to explore the design space, understand how different parameters relate to each other, areas of feasible and unfeasible designs, etc.

Documentation

A detailed documentation of the aixd library is provided here. The documentation includes detailed installation instructions, API references, a user guide, application examples and more.

Installation

Requirements

• Python >= 3.9

The aixd library can be installed using uv, pip or conda. The recommended way is to use uv for its automatic PyTorch backend selection, but you can also use pip or conda if you prefer.

Note: It is recommended to use virtual environments to manage the dependencies of your projects.

Using conda

Create a new environment and install aixd:

conda create -n venv python=3.9
conda activate venv
conda install -c conda-forge aixd

Using pip

Create a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

Install aixd:

pip install aixd

Using uv

Create a virtual environment:

uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

Install aixd with automatic PyTorch backend selection:

uv pip install aixd --torch-backend=auto

The --torch-backend flag automatically detects your CUDA driver, AMD ROCm driver, or Intel GPU and selects the most suitable PyTorch version. If no compatible GPU is found, the CPU version will be installed. For more information, refer to the uv documentation.

Latest development version

Install the latest version using pip from the git repository:

pip install --upgrade git+https://gitlab.datascience.ch/ai-augmented-design/aixd.git

Examples

The aixd toolbox includes a variety of example applications to showcase its features and usage. These examples are located in the examples folder and cover different design tasks, each with self-contained Jupyter notebooks and detailed explanations.

For more information about each example, refer to the corresponding README.md file within each example folder.

To run the examples, ensure you have downloaded the examples folder and installed the aixd library with all example dependencies:

pip install "aixd[examples]"

Development

If you are going to develop on this repository, perform an installation from source:

git clone https://gitlab.datascience.ch/ai-augmented-design/aixd.git
cd aixd

Create and activate a python virtual environment. Install all dependencies including the examples and development tools.

Using conda:

conda env create -f environment.yml

Or using pip:

pip install -e ".[examples, dev]"

Or using uv (recommended for proper torch backend selection):

uv pip install -e ".[examples, dev]" --torch-backend=auto

Check the contribution guidelines for more details.

Folders and structure

The structure we follow on the current repo is as follows:

• examples : all example applications of the aixd toolbox
• src : for all source code. It can be structure following the next structures
  • src/aixd : source code of aixd toolbox

Known issues

• Plotly image export can cause a hang of the system. This is due to a bug in Kaleido (the library used by Plotly for image export) reported in here. A workaround is to downgrade Kaleido to version 0.1.0.post1, which can be done by running pip install kaleido==0.1.0.post1.