Pycsou-gsp is the graph signal processing extension of the Python 3 package Pycsou for solving linear inverse problems. The extension offers implementations of graph convolution and differential operators, compatible with Pycsou’s interface for linear operators. Such tools can be useful when solving linear inverse problems involving signals defined on non Euclidean discrete manifolds.

Graphs in Pycsou-gsp are instances from the class pygsp.graphs.Graph from the pygsp library for graph signal processing with Python.

Content

The package is organised as follows:

1. The subpackage pycgsp.linop implements the following common graph linear operators:

• Graph convolution operators (pycgsp.linop.conv):

pycgsp.linop.conv.GraphConvolution(Graph, …)

Graph convolution.

• Graph differential operators (pycgsp.linop.diff):

pycgsp.linop.diff.GraphLaplacian(Graph[, dtype])	Graph Laplacian.
pycgsp.linop.diff.GraphGradient(Graph[, dtype])	Graph gradient.
pycgsp.linop.diff.GeneralisedGraphLaplacian(Graph)	Generalised graph Laplacian operator.

2. The module pycgsp.graph provides routines for generating graphs from discrete tessellations of continuous manifolds such as the sphere:

pycgsp.graph.cvxhull_graph(R[, …])	Build the convex hull graph of a point set in \(\mathbb{R}^3\).
pycgsp.graph.healpix_nngraph(nside[, …])	Build the nearest neighbour graph of a HEALPix spherical point set.