PERF: asv for import time

[jbrockmendel]

Not entirely sure how to do it with asv, but at the command line its just python3 -X importtime -c "import pandas"

• import time with python3 -X importtime -c "import pandas"
• import size with something like import pandas, psutil; psutil.Process().memory_info()

[WillAyd]

Yea I think this would be a good idea. Maybe create a subprocess in setup and load pandas in the subprocess call

[jbrockmendel]

airspeed-velocity/asv#487

[alimcmaster1]

Here is something similar numpy did: https://github.com/numpy/numpy/pull/8159/files - do we just want to benchmark import pandas or test some sub packages too?

[WillAyd]

Makes sense to me. I think we'd take a PR for something like that if you'd like to post one

[TomAugspurger]

Ideally, the setup would ensure that matplotlib and numexpr (really, all our optional deps) aren’t in the Python env running the benchmark.

…

On Jun 8, 2019, at 15:50, William Ayd ***@***.***> wrote: Makes sense to me. I think we'd take a PR for something like that if you'd like to post one — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[jbrockmendel]

No opinion on numexpr, but for matplotlib I'd rather it be installed. I'm pretty sure some effort has been put in to making matplotlib imports lazy. This would help ensure that we don't accidentally undo that (this was motivated by exactly this happening in statsmodels)

[TomAugspurger]

Matplotlib has to be imported by default right now because we register formatters. That’s deprecated, and then we can ensure it’s lazily imported. We have a separate CI check ensuring that certain packages aren’t imported with pandas (search for import time)

…

On Jun 9, 2019, at 10:33, jbrockmendel ***@***.***> wrote: No opinion on numexpr, but for matplotlib I'd rather it be installed. I'm pretty sure some effort has been put in to making matplotlib imports lazy. This would help ensure that we don't accidentally undo that (this was motivated by exactly this happening in statsmodels) — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[TomAugspurger]

pandas/ci/code_checks.sh

Line 192 in c7748ca

if [[ -z "$CHECK" || "$CHECK" == "code" ]]; then

[jbrockmendel]

Thanks for pointing that out. Is there a particular reason that was used instead of the approach here statsmodels/statsmodels#5831 ?

[TomAugspurger]

You mean a separate check, not run as part of the pytest test suite? Probably no reason; seems easier to me that it be separate since it isn’t really a traditional unit test.

…

On Jun 9, 2019, at 11:43, jbrockmendel ***@***.***> wrote: Thanks for pointing that out. Is there a particular reason that was used instead of the approach here statsmodels/statsmodels#5831 ? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.