tools.lrmatrix

Contents

tools.lrmatrix#

bartorch.tools.lrmatrix()#

Perform (multi-scale) low rank matrix completion

Runs bart lrmatrix.

Parameters:

  • input (torch.Tensor) – Input array.

  • d (bool, default=False) – perform decomposition instead, ie fully sampled (-d)

  • i (int | None, default=None) – maximum iterations. (-i)

  • m (int | tuple[int, ...] | None, default=None) – Axes reshaped into the matrix columns. (-m)

  • f (int | tuple[int, ...] | None, default=None) – Axes the multi-scale partition is along. (-f)

  • j (int | None, default=None) – block size scaling from one scale to the next one. (-j)

  • k (int | None, default=None) – smallest block size (-k)

  • N (bool, default=False) – add noise scale to account for Gaussian noise. (-N)

  • s (bool, default=False) – perform low rank + sparse matrix completion. (-s)

  • l (int | None, default=None) – perform locally low rank soft thresholding with specified block size. (-l)

  • o (str | None, default=None) – summed over all non-noise scales to create a denoised output. (-o)

  • u (bool, default=False) – () (-u)

  • v (bool, default=False) – () (-v)

  • H (bool, default=False) – (hogwild) (-H)

  • p (float | None, default=None) – (rho) (-p)

  • n (bool, default=False) – (no randshift) (-n)

  • **extra (Any) – Further BART flags, passed through by name.

Returns:

output

Return type:

torch.Tensor

Contents