Add MESMER-M integration tests #501
Conversation
for more information, see https://pre-commit.ci
|
Thanks - can you save the coeffs as netCDF and not pickle (I know it's a bit more annoying but we want to move away from pickle) |
veni-vidi-vici-dormivi
force-pushed
the
m_integrationtests
branch
from
17afff8
to
f97ea6e
Compare
for more information, see https://pre-commit.ci
|
I haven't looked into it so not sure what is going on but a bit strange that none of the environments get it right. I thought this does not have to be high priority but then it would be good to stabilize the numerics for memser-m. What helped for the covariance thingy was testing the code on cfc vs exo, which have different cpus with different sets of SIMD instructions. But I am not really working today - if I find time I'll take a look tomorrow or else after our retreat... |
|
Okay I think (for the moment) this is good enough. I can open an issue on this after merging so we can come back to it maybe in the future. Also, should I separate the changes in the harmonic model and the power transformer from the integration tests into (3) separate PRs or just rename this PR to "MESMER-M integration test stabilizing"? |
| AR1_fit.residuals, weights, phi_gc_localizer, "time", 30 | ||
| ) | ||
|
|
||
| # merge into one dataset |
There was a problem hiding this comment.
I am not a huge fan of this there and back renaming and merging. Would it be bad to save each Dataset as individual netCDF? Or merge into a datatree? (But maybe later).
There was a problem hiding this comment.
I like the idea of having one dataset with all the calibrated parameters per ESM. I did some renaming in the modules such that everything can be merged without problems. What do you think about that?
|
Do you happen to know if the estimates are stable for each OS? Could we do os-dependent tolerances? (Not asking you to sink more time into this PR atm...) |
CHANGELOG.rst