Installation

Install from PyPI

Install the core edge2torch package from PyPI with:

pip install edge2torch

The core installation supports compiling sparse PyTorch models from edgelists, aligning named input features, customizing compiled models, and training with ordinary PyTorch.

Optional interpretation support

interpret_model() uses Captum and is installed as an optional dependency.

To install edge2torch with interpretation support:

pip install "edge2torch[interpret]"

Optional AnnData support

For optional AnnData input support:

pip install "edge2torch[anndata]"

To install both interpretation and AnnData support:

pip install "edge2torch[all]"

Development installation

To work on the package locally, clone the repository and install it in editable mode from the project root:

git clone git@github.com:Thomas-Rauter/edge2torch.git
cd edge2torch
pip install -e .

For optional interpretation support during development:

pip install -e ".[interpret]"

For optional AnnData support during development:

pip install -e ".[anndata]"

For both optional interpretation and AnnData support:

pip install -e ".[all]"

Optional dependency groups

Install development dependencies with:

pip install -e ".[dev]"

Install documentation dependencies with:

pip install -e ".[docs]"

Notebook and documentation note

Some documentation notebooks use optional visualization tools such as Graphviz. If a notebook requires Graphviz rendering, you may also need the system-level Graphviz installation in addition to the Python package.

For example, on Ubuntu or Debian:

sudo apt install graphviz

Verify the installation

A minimal core-installation smoke test is:

python -c "import edge2torch; print('edge2torch imported successfully')"

To verify interpretation support, install edge2torch[interpret] and run:

python -c "from edge2torch import interpret_model; print(interpret_model)"