gsxform¶

Wavelet scattering transforms on graphs via PyTorch

gsxform is a package for constructing graph scattering transforms, leveraging PyTorch to allow for GPU based computation. Using PyTorch, gsxform offers the ability to more easily build models that use both scattering transform and neural network components.

gsxform is first and foremost a research project and is being continuously refined. Behavior can potentially be unstable and consistency is not guaranteed.

Installation¶

gsxform requires Python 3.10 or later.

Latest version (recommended)¶

uv add "gsxform @ git+https://github.com/armaank/gsxform"

Or, to use a specific branch not yet merged to main

uv add "gsxform @ git+https://github.com/armaank/gsxform@branch-name"

Stable Version¶

To install the latest stable release from PyPI:

uv add gsxform
# or
pip install gsxform

Development version¶

gsxform uses uv to manage environments and dependencies. To work on gsxform itself, clone the repository and run make install:

git clone https://github.com/armaank/gsxform.git
cd gsxform
make install

This creates the virtual environment, installs gsxform with development and documentation dependencies, and sets up the pre-commit hooks. The development guide covers the make targets, code style, branching, and the release process.

License¶

The original code of this repository is released under the BSD 3.0-Clause Licence. Modifications, adaptations and derivative work is encouraged!

Citation¶

If you use gsxform, please cite using the Zenodo DOI