domfournier

It also looks like `mkl=2021.2.0` is incompatible with `pymatsolver`  Recommend downgrading to 2020.4 **conda install -c conda-forge mkl=2020.4** 

Fun fact, the `mkl_get_max_threads` returns the number of cores on the machine, which is half the thread count for dual-threaded systems...

Oh yea totally, I recall now. The dask operation could do it in place, but (maybe) not numpy. Let's leave this guy open for a little while as a reminder.

Wow that was quick. Thanks Einsten (@jcapriot ;) Just tried on this: `import numpy as np G = np.random.randn(10000,100000) # 7.4 Gb matrix W = np.ones(G.shape[0])` This indeed works: `np.einsum('i,ij,ij->j',...

wowzer, this needs to be updated. I think what you did would, but we need to fix it on our end for sure.

Hi @IMirzalizade, glad you found us. So these parameters belong to different objects, so they obviously have very different meanings. All very good questions though. ``reg.norms``: Used by `Regularization.Sparse` to...

I agree, we just need to specify what type of survey we are expecting on the read, and return nan on the elevations if not included. The issue is more...

Technically it wouldn't matter, but iterative solvers can be very slow to converge if the system is poorly conditioned. Hence why we added a property `approxHinv` to the `ProjectedGNCG` and...

This looks like amazing work @ASmelov ! Two things to look out for: 1) for bulk physical property model like these you need to make sure your domain extends much...

[From Slack] Increasing the domain size outside of the core can be done pretty efficiently with `padding cells`. The rule of thumb is to have padding cells on either side...