SpectralMatching

class src.qap_solvers.spectral_matching.SpectralMatching(max_iter=50, stop_thresh=2e-07)

Spectral Graph Matching solver. Also known as Power Iteration layer, which computes the leading eigenvector of input matrix. For every iteration,

v_k+1 = M * v_k / ||M * v_k||_2

Parameter: maximum iteration max_iter Input: input matrix M

(optional) initialization vector v0. If not specified, v0 will be initialized with all 1.

Output: computed eigenvector v

forward(M, v0=None, **kwargs)

training: bool