cr.sparse.block.bsbl.bsbl_em_jit

cr.sparse.block.bsbl.bsbl_em_jit(Phi, y, blk_len, options=BSBL_Options(learn_block_corr=1, learn_lambda=1, prune_gamma=0.001, lambda_val=1e-12, max_iters=800, epsilon=1e-08))

Reconstructs a block sparse signal using BSBL-EM algorithm

Parameters

  • Phi (jax.numpy.ndarray) – Sensing matrix

  • y (jax.numpy.ndarray) – Measurement vector

  • blk_len (int) – Length/size of each block

  • options (BSBL_Options) – Options for algorithm execution

Returns

Solution of the sparse recovery problem

Return type

BSBL_State

Note

  • Phi must be a matrix. Linear operators are not supported since we need to break Phi down into submatrices for each block.

  • Use bsbl_em_options() to initialize options for the algorithm.

Examples