search

astro-informatics / ssht

Fast and exact spin spherical harmonic transforms
http://astro-informatics.github.io/ssht/
GNU General Public License v3.0
11 stars 2 forks source link

Draft: add ducc support for forward adjoint #58

Closed mreineck closed 3 years ago

mreineck commented 3 years ago

This PR adds all necessary machinery to allow forward_adjoint operations with ducc0.

Problem: ssht and ducc0 currently produce different results when running forward on non-bandlimited maps, and so it is not entirely clear how to test the consistency between both implementations. We'll have to iterate on this a bit, I think.

jasonmcewen commented 3 years ago

Just linking to #55 since some of the discussion there relates to this PR.

jasonmcewen commented 3 years ago

@mdavezac Any idea why tests are failing here?

mreineck commented 3 years ago

The tests fail because the released version of ducc0 does not have the forward_adjoint method yet. I'll update this during the next days.

mreineck commented 3 years ago

I have created a new ducc0 release with which the tests should work.

The fully general adjointness test for the forward/forward_adjoint pair still fails (locally for me, I have not added it to the test suite), but the tests equivalent to those in ssht itself should pass.

mreineck commented 3 years ago

CI looks OK, but please double check that I'm really testing the things you are interested in!

jasonmcewen commented 3 years ago

Thanks @mreineck, sounds good. I think the tests you have are sufficient and the extra one (i.e. the forward/forward_adjoint pair) is not necessary (as discussed further in #55) .

@mdavezac Could you take a look when you get a moment and let us know if you've any comments? If not we can go ahead and merge (and create a new release).

mdavezac commented 3 years ago

Looks good. I'll just add a few :lipstick: commits, bump the minor version, and then I'll merge.