Chebyshev Convolution

class ChebyConvolve(L)

Bases: object

Chebyshev polynomial graph convolution context.

Approximates spectral graph filters using Chebyshev polynomials of the first kind via the recurrence relation.

Parameters:

L (csc_matrix) – Sparse Graph Laplacian.

convolve(B, C)

Performs graph convolution using Chebyshev polynomial approximation.

Parameters:

• B (ndarray) – Input signal of shape (n_vertices,) or (n_vertices, n_signals).

• C (ChebyKernel) – Chebyshev kernel with coefficients and spectral bound.

Returns:

Convolved signal of shape (n_vertices, n_signals, n_dims) or (n_vertices, n_dims) for 1D input.

Return type:

ndarray

convolve_multi(B, kernels)

Apply multiple kernels efficiently by sharing Chebyshev term computation.

Parameters:

• B (ndarray) – Input signal.

• kernels (list of ChebyKernel) – Kernels to apply.

Returns:

Convolved signals for each kernel.

Return type:

list of ndarray