CoSaMP

Compressive sampling matching pursuit (CoSaMP) algorithm.

This implementation currently does not handle complex signals.

mr_utils.cs.greedy.cosamp.cosamp(A, y, k, lstsq='exact', tol=1e-08, maxiter=500, x=None, disp=False)[source]

Compressive sampling matching pursuit (CoSaMP) algorithm.

Parameters:
  • A (array_like) – Measurement matrix.
  • y (array_like) – Measurements (i.e., y = Ax).
  • k (int) – Number of expected nonzero coefficients.
  • lstsq ({'exact', 'lm', 'gd'}, optional) – How to solve intermediate least squares problem.
  • tol (float, optional) – Stopping criteria.
  • maxiter (int, optional) – Maximum number of iterations.
  • x (array_like, optional) – True signal we are trying to estimate.
  • disp (bool, optional) – Whether or not to display iterations.
Returns:

x_hat – Estimate of x.
Return type:

array_like

Notes

lstsq function - ‘exact’ solves it using numpy’s linalg.lstsq method. - ‘lm’ uses solves with the Levenberg-Marquardt algorithm. - ‘gd’ uses 3 iterations of a gradient descent solver.

Implements Algorithm 8.7 from [1].

References

[1]Eldar, Yonina C., and Gitta Kutyniok, eds. Compressed sensing: theory and applications. Cambridge University Press, 2012.