brandon-rhodes
commented
1 year ago The slow Python-language model.py file is only used when sgp4 failed to compile the actual C++ SGP4 implementation when it installed on your machine. And since the SGP4 implementation cannot be vectorized (at least not without being fairly completely rewritten), both the Python and C++ versions of the code do call the underlying sgp4() routine in a loop; for Python it's, of course, a slow loop, and in C it's a very fast one.
I will make two improvements to the documentation to resolve this issue:
- I will document the ability to
print(sgp4.api.accelerated)to see whether the C++ compiled and will make computations faster. I suggest you try running that, and sharing the results here on this issue, so that we can confirm that you don't have the C++ module available. - I will add a paragraph to the "Array Acceleration" section of the docs that mentions that arrays really only achieve a noticeable speedup if the C++ version compiled, and that otherwise the loop will run slowly.
Thanks for bringing this problem with the docs to my attention!
If you can find the errors that pip encountered when installing sgp4, we might be able to track down why you don't have the C++ compiled. Maybe you can find the pip install logs still sitting around?
jeslinmx
Let me first start off by saying that I love this library and Skyfield which it supports; it's been an incredibly useful and essential tool in my work.
I noted that in the documentation, the stated purpose of
Satrec.sgp4_arrayandSatrecArray.sgp4is "to avoid the expense of Python loops when you have many" dates and/or satellites. Accordingly, one would expect that the array of dates to be handled in a vectorized manner, a la Numpy.However, it does not seems that these 2 methods are just wrapping
Satrec.sgp4in a python loop in the backend: https://github.com/brandon-rhodes/python-sgp4/blob/8f44082e3b16cc40f3a170b5ed819919a334c6f2/sgp4/model.py#L129-L131 https://github.com/brandon-rhodes/python-sgp4/blob/8f44082e3b16cc40f3a170b5ed819919a334c6f2/sgp4/model.py#L169-L172This also matches with what I observe in SnakeViz when using Skyfield (find_events of multiple satellites[^1], FWIW):
(note how 767 calls of
sgp4_arrayresults in 36658755 calls ofsgp4)Is this the intended behavior, or a work in progress? Or am I misunderstanding the issue/incorrectly using Skyfield?
[^1]: on that note, is there possible to declare multiple satellites (say, a constellation) as a single object in Skyfield, and have their propagation handled in the backend by a
SatrecArrayrather than individualSatrecs?