Swap to adam_core's Propagator, Orbits, and Observers

[moeyensj]

Replaces:

• Backend class with adam_core's Propagator class
• Orbits with adam_core's Orbits
• Observers with adam_core's Observers
• coordinate transformations with adam_core's transformations

Removes:

• Old implementation of range and shift
• Data pre-processing function that will not be necessary with the new typed structure of Observations

All unit tests pass, but most importantly:
pytest -k ivezic -m integration passes! This tests the entire pipeline.

The benchmark shows a slowdown from 1.2695 to 1.4256 seconds. This is more than likely due to the conversion from adam_core types to dataframes still used by a large part of the THOR pipeline. I'll change those types out in a follow up PR.

[akoumjian]

Just a quick read-through. So nice to clean this up!

[akoumjian]

Why do we use the private method here?

[moeyensj]

No good reason, we were calling generate_ephemeris (original function) with num_jobs=1 which would have called the private function in series. I'll go through and replace the private calls throughout IOD, OD, and the arc extension code.

[akoumjian]

Let's think about how to make this open enough that people can install third-party packages and use other propagators.

[moeyensj]

Yes, good idea. In range and transform we pass the actual class, not a string. Maybe we pass the class all the way through to the worker functions and then initialize it in there?

propagator: Type["Propagator"] = PYOORB,
propagator_kwargs: dict = {},

Then inside worker functions:

prop = propagator(**propagator_kwargs)

[moeyensj]

I'll clean all of this up once I have the quivr types built out and can work on the parallelization again.

[akoumjian]

Yeah, that makes sense to me.

…

On Thu, Nov 2, 2023 at 09:48 Joachim Moeyens ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In thor/orbits/iod.py <#126 (comment)>: > @@ -178,8 +184,8 @@ def iod( observation_selection_method="combinations", iterate=False, light_time=True, - backend="PYOORB", - backend_kwargs={}, + propagator: Literal["PYOORB"] = "PYOORB", Yes, good idea. In range and transform we pass the actual class, not a string. Maybe we pass the class all the way through to the worker functions and then initialize it in there? propagator: Type["Propagator"] = PYOORB,propagator_kwargs: dict = {}, Then inside worker functions: prop = propagator(**propagator_kwargs) — Reply to this email directly, view it on GitHub <#126 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFTGJF3TLUV3CSQQ3VUAGLYCOQEPAVCNFSM6AAAAAA62JRC3KVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMYTOMJQGM2DOMBWGI> . You are receiving this because you commented.Message ID: ***@***.***>