SkyCoord slow to import

[sdjohnson-astro]

I recently wrote a few command line tools for converting coordinates but found that they are noticeably slow to run. The runtime is dominated by importing SkyCoord from astropy.coordinates which takes nearly 2 seconds (for reference, this is a new installation of astroconda running on my MacBook Pro retina w/ 2.7 GHz i7 processor, 16 GB of ram, and an SSD). For comparison, imports from standard libraries such as numpy or UnivariateSpline from scipy.interpolate takes ~0.15 seconds. Is there any way to speed up the import of SkyCoord?

[MSeifert04]

@sdjohnson-astro Which version of astropy do you use? The import time was sped up for astropy 1.2 but it's still not as fast as numpy/scipy.

See also #4598.

[sdjohnson-astro]

I am using Astropy version 1.2.1

Thanks for the super fast reply!

[pllim]

Like @MSeifert04 said, there is already a related issue like this. What if you simply do import astropy? Is it the same kind of slowness?

[sdjohnson-astro]

import astropy takes 0.5 seconds. Significantly faster than the 2 seconds it takes to import SkyCoord from astropy.

[pllim]

If you are feeling adventurous, you can profile the offending import statement and post some results here. See https://docs.python.org/3/library/profile.html

[pllim]

And to make sure that this is not already solved since the release, you can also try the dev version from master.

[astrofrog]

@sdjohnson-astro - can you test import astropy.units?

[MSeifert04]

Just for convenience these import times can be tested from the command line with
python -m timeit -n 1 -r 1 "from astropy.coordinates import SkyCoord" (which takes 5-6s on my machine)

[pllim]

Interesting. Using the latest dev myself. The first time I run @MSeifert04 's command, it took 4.83 sec. But subsequent calls was about 1.87 sec each.

[pllim]

@astrofrog , python -m timeit -n 1 -r 1 "import astropy.units" gave me about 925 msec. But that is after I already ran the SkyCoord import test several times.

[sdjohnson-astro]

Thanks for the tip.

@astrofrog importing astropy.units takes 0.8 seconds on my laptop.

I don't know enough about astropy to make much of the profiling results at first glance. The profile results are attached if anyone wants to take a look.
SkyCoord_profile.txt

I will try out the latest dev myself, but from @pllim it looks like the import time is still non-negligible.

[MSeifert04]

@pllim The first time python encounteres new .py files it generates __pycache__ and .pyc files. This will account for the longish "first" import.

[pllim]

@sdjohnson-astro , thanks for the log. But it is hard to diagnose without full path to the filenames, and would be nice to sort them by cumtime.

[sdjohnson-astro]

@pllim A version sorted by cumulative time is attached. I am not sure how to get the full path to filenames with cProfile. Is there an option in the documentation that I am missing?

Thanks!

import_SkyCoord_profile.txt

[pllim]

Thanks, @sdjohnson-astro . I think the listed filenames are unique enough that we don't need the full path. From your log:

        1    0.001    0.001    1.505    1.505 baseframe.py:6(<module>)
        1    0.000    0.000    1.500    1.500 earth.py:2(<module>)
        1    0.003    0.003    1.499    1.499 __init__.py:25(<module>)
        1    0.000    0.000    1.448    1.448 fk5.py:3(<module>)
        1    0.000    0.000    1.418    1.418 utils.py:6(<module>)
        1    0.001    0.001    1.417    1.417 iers.py:10(<module>)

Maybe @eteq or @taldcroft wish to chime in?

[sdjohnson-astro]

@pllim Importing SkyCoord also takes 2 seconds on my computer when using the latest 'dev'.

[pllim]

@sdjohnson-astro , thanks for checking! I see iers.py listed. I wonder if it is something to do with some kind of initialization there...

[eerovaher]

I ran the following with the current astropy development version (0e379d8):

python -X importtime -c "from astropy.coordinates import SkyCoord"
python -X importtime -c "import numpy"

The difference in the reported times was roughly a factor of 3, much less than the more than an order of magnitude difference reported in this issue. I haven't bothered looking up any commits that might have contributed towards faster importing, but it does seem like this issue is no longer relevant.

[sdjohnson-astro]

This is a blast from the past! Yes, my understanding is that a decent amount of effort went into speeding things up under the hood some time ago.

…

----------------------------------- Sean D. Johnson (he/him) Assistant Professor & LSA Collegiate Fellow Department of Astronomy University of Michigan ***@***.***

On Nov 15, 2022, at 6:07 PM, Eero Vaher ***@***.***> wrote: I ran the following with the current astropy development version (0e379d8 <0e379d8>): python -X importtime -c "from astropy.coordinates import SkyCoord" python -X importtime -c "import numpy" The difference in the reported times was roughly a factor of 3, much less than the more than an order of magnitude difference reported in this issue. I haven't bothered looking up any commits that might have contributed towards faster importing, but it does seem like this issue is no longer relevant. — Reply to this email directly, view it on GitHub <#5222 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AE7MYXOARGQZNDUM5PQROHLWIQJRRANCNFSM4CL2K7OQ>. You are receiving this because you were mentioned.

[github-actions]

Hi humans 👋 - this issue was labeled as Close? approximately 6 hours ago. If you think this issue should not be closed, a maintainer should remove the Close? label - otherwise, I will close this issue in 7 days.

If you believe I commented on this issue incorrectly, please report this here

[github-actions]

I'm going to close this issue as per my previous message, but if you feel that this issue should stay open, then feel free to re-open and remove the Close? label.

If this is the first time I am commenting on this issue, or if you believe I closed this issue incorrectly, please report this here