bartorch.optim.maxeigen

Contents

bartorch.optim.maxeigen#

bartorch.optim.maxeigen(A, terms=None, *, cclambda=0.0, iterations=30)#

BART’s estimate of the largest eigenvalue of the operator a step divides by.

What pics -e asks for. It is a power iteration from a random start, so it draws on BART’s own generator: a loop written outside the library has to ask for it here, at the point in the sequence the library would have asked, or the draws that follow – a wavelet term’s cycle spinning, say – are different ones.

Parameters:

  • A (LinearOperator) – The encoding. Its normal is the operator, with cclambda on the diagonal, as lsqr builds it.

  • terms (Regularizer or iterable of Regularizer, optional) – Terms whose transforms are added to it. That is what the primal-dual iteration estimates over; the proximal ones take the encoding alone.

  • cclambda (float) – The quadratic weight (pics -q).

  • iterations (int) – Power iterations; BART takes thirty.

Return type:

float

Contents