domfournier
domfournier
I would vote that we keep the implementation in line with the other EM codes that have been parallelized already (i.e. FEM and DC). So the factorization of A.T is...
Well if you are going to store J (either in RAM or disk), then you technically don't need A.T ever until you update the model. So no point in having...
I would say yes to everything above. Ideally, though, each computational "chunk" should have to only worry about one ATinv. For frequency that's easy (one misfit per frequency). For time-domain...
Might be a seperate issue, but it would be very beneficial to re-work the way fields are stored, mostly trying to remove the `mesh` object as much as possible. Mostly...
All right boys, finishing the year on a high. First implementation of storing J for TEM here: https://github.com/MiraGeoscience/simpeg/blob/dask_tem/SimPEG/dask/electromagnetics/time_domain/simulation.py @dccowan Checkout the compute_J. I combined the two for-loops over time to...
It's a bit stale, but I would like to meet with John W before merging or closing. If his approach works better, then there we would no point in bringing...
@jcapriot Yea I saw your modifications to the UpdateSensWeights. I like yours better. Looks like we may be able to share some components indeed. Not sure how you want to...
@jcapriot I refactored using your base class.
@jcapriot Ready for another review.
ready for review @simpeg/simpeg-developers