PowerMethod will produce slightly different results for every call

[evelinaametova]

https://github.com/vais-ral/CCPi-Framework/blob/598f758d660e9da92878d6819502321cb7c2b09e/Wrappers/Python/cil/optimisation/operators/Operator.py#L134

Shouldn't this random allocation be seeded? If not seeded, reconstructions with exactly the same input will produce slightly different results.

[jakobsj]

Yes, should be seeded for reproducibility.

On Mon, 22 Mar 2021 at 09:03, evelinaametova @.***> wrote:

Assigned #806 https://github.com/vais-ral/CCPi-Framework/issues/806 to @jakobsj https://github.com/jakobsj.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/vais-ral/CCPi-Framework/issues/806#event-4488498790, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACMDSCHBBH7HSU3DYHSEHE3TE32VJANCNFSM4ZSTJEHA .

[paskino]

Suggest to add a new self._random_number_seed member in LinearOperator that will be used by calculate_norm https://github.com/TomographicImaging/CIL/blob/3e29c5b512485e563935053b33b7dba5d9187690/Wrappers/Python/cil/optimisation/operators/Operator.py#L154-L167

to create x0 if no x_init is passed

[MargaretDuff]

In https://github.com/TomographicImaging/CIL/pull/1645 we found that we needed a random seed for the power method tests! It would be good to be able to set the random seed in the argument of the power method

[paskino]

We discussed on 6/2/25 that we could also implement the calculation of the norm of the operator with the Jacobi iterations as implemented by @evgueni-ovtchinnikov in SIRF. The method converges much quicker than the PowerMethod. Here how that is called.