Install K GraphFuse¶

Requirements¶

Component	Requirement	Purpose
Python	3.11 or newer	Package and reference tests
NumPy	1.26 or newer	Channel coefficients and dense feature states
SciPy	1.11 or newer	Sparse graph operators
PyTorch	2.11 or newer, CUDA build	Optional GPU execution
NVIDIA device and driver	Compatible with your PyTorch CUDA build	Optional GPU execution

CPU import does not import PyTorch. cuGraph, CuPy, Java and Graphalytics data are comparator requirements only; they are not required to use the library. Support is limited to the tested combinations described in the support matrix.

Install from the repository¶

git clone https://github.com/Utkarsh-Sinha0/k-graphfuse.git
cd k-graphfuse
python -m venv .venv

Activate the environment on Linux or macOS:

source .venv/bin/activate

Activate the environment on Windows PowerShell:

.venv\Scripts\Activate.ps1

Then install and verify:

python -m pip install .
python examples/quickstart.py
python examples/sensitivity.py
python examples/resolvent.py

Expected quickstart result: retained degree 2, output [5.125 5. ].

Enable CUDA¶

Use the official PyTorch installation selector to choose a build compatible with your operating system and driver. Then install the optional runtime and verify device availability:

python -m pip install ".[gpu]"
python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"
python examples/gpu.py

True confirms that PyTorch can access a CUDA device. A package installation alone does not confirm GPU support. Follow the CUDA execution guide before choosing FP32 or CUDA capture.

Develop locally¶

python -m pip install -e ".[docs]"
python -m unittest discover -s tests -v
python -m mkdocs serve

The site preview defaults to http://127.0.0.1:8000/. Run python -m mkdocs build --strict for publication checks. CUDA tests report explicit skips on machines without CUDA; the CPU tests still run.