(don't merge) Vectorize meshes #391
Conversation
solve_mcmc() now checks mcmc_station_velocity_method before deciding whether to use streaming mode (discard_tde_to_velocities=True). Only project_to_eigen supports streaming; direct and low_rank require the full tde_to_velocities matrices. Also adds NotImplementedError in _station_vel_from_elastic_mesh() when direct/low_rank is used with operators that lack tde_to_velocities.
|
@brendanjmeade just adding context here that we've made the modified model for big WNA pictured (in simplified form) above. It can be fully loaded in, compiled, and sample, but Adrian is working on rewriting it into what will no doubt be far simpler and efficient than my initial, rough-sketch implementation :) |
|
@elc45 @aseyboldt Implementing a good idea in a "far simpler and efficient" manner is time well spent! |
|
Note Heads up, @elc45! This PR doesn't modify Remote git checkout tde-optional
git fetch upstream
git merge upstream/mainMerge-base: 7ba8807, latest lockfile change on This notice will be minimized automatically once Generated automatically by |
3 participants
@aseyboldt something like this to take the model roughly from
to
I advocate for removing all tde-level things from the model altogether and just rebuild everything post hoc. Except for in the coupling-constrained case I guess we need to construct the elastic fields to do coupling_coefs -> coupling_field -> elastic_field -> velocities. I'd also propose we finish Matern before doing this