Estimating L0 parameter in AcceleratedPGM about scico

[bwohlberg]

The error is because you used C.H * C rather than C.H @ C. But you don't need to implement the estimation yourself -- see scico.linop.operator_norm.

[shnaqvi]

Oh nice, so this is actually computing the sqrt. of the largest eigenvalue of the Hessian of the operator! Thanks @bwohlberg for pointing out my mistake.

Can you also please shed a light on how can I use PGMStepSize in lieu of estimating L0 beforehand using the operator_norm as it is taking really long to run on my sum of weighted convolutions blur operator?

[bwohlberg]

You should find some relevant examples here.

[shnaqvi]

Thanks @bwohlberg , I was able to get clarity on using the step size from the above example.