LION.classical_algorithms.spgl1_torch¶

LION.classical_algorithms.spgl1_torch(op, y, **spgl1_kwargs)[source]¶

Solves an l1 sparse reconstruction by SPGL1 (van den Berg and Friedlander, 2008), the spectral projected-gradient method of the spgl1 package, with op (a linear operator, called as op(w) and op.adjoint(r)) for the products. y: the measurements (M,); other keyword arguments go to spgl1.spgl1 as they are (its tolerances, iterations, …; https://spgl1.readthedocs.io). Returns the coefficients, with the shape of op.adjoint(y).

References: van den Berg and Friedlander, “Probing the Pareto frontier for basis pursuit solutions”, SIAM J. Scientific Computing 31(2):890-912, 2008; “Sparse optimisation with least-squares constraints”, TR-2010-02, University of British Columbia, 2010.