GraphDiffusion is a python package based on the DiffusionGeometry library in Matlab (https://services.math.duke.edu/~mauro/code.html#DiffusionGeom).
-
GraphDiffusion has been implemented in Python3 and can be installed using
$> git clone [email protected]:pkathail/GraphDiffusion.git $> cd GraphDiffusion $> sudo pip3 install . -
GraphDiffusion depends on a number of
python3packages available on pypi and these dependencies are listed insetup.pyAll the dependencies will be automatrically installed using the above commands
After installation, GraphDiffusion can be used with the following commands
$> import GraphDiffusion
$> res = GraphDiffusion.graph_diffusion.run_diffusion_map(data, knn=10, normalization='smarkov')
where data is a N x D matrix representing N points in R ^ D, knn is the number of nearest neighbors, normalization is the method for normalizing weights. Please refer to the docstring for more details.
res is dictionary with the following objects:
T:N x Nsparse matrix giving the normalized diffusion operatorW:N x Nsparse matrix of weightsEigenVectors: Eigen vectors of matrixTEigenValues: Eigen values of matrixT