search

alkaline-ml / pmdarima

A statistical library designed to fill the void in Python's time series analysis capabilities, including the equivalent of R's auto.arima function.
https://www.alkaline-ml.com/pmdarima
MIT License
1.57k stars 231 forks source link

[MRG+1] Add aarch64 wheel build support #435

Closed odidev closed 2 years ago

odidev commented 3 years ago

Added aarch64 wheel build support. Related to https://github.com/alkaline-ml/pmdarima/issues/434, @aaronreidsmith Could you please review this PR?

aaronreidsmith commented 3 years ago

Thanks for opening #435 and this PR! I think we'll want to do a little more testing with this architecture before approving/merging this. Additionally, just to maximize our parallelism, GitHub Actions may not be the place for it to run.

This PR also reverts our numpy changes that were necessary to support sktime (see #423, #425, https://github.com/alan-turing-institute/sktime/pull/828, and https://github.com/alan-turing-institute/sktime/pull/831 for context), so we will need to test with our supported version of numpy. Tagging @tgsmith61591 as well to move the conversation forward.

odidev commented 3 years ago

@tgsmith61591, Could you please have a look at this PR? Thanks in advance.

odidev commented 3 years ago

Done.

aaronreidsmith commented 3 years ago

Hey @odidev, sorry to give you the run around for so long on this PR. It looks like our CI keeps timing out trying to build statsmodels from scratch for aarch64. This seems to be a known issue that they are working on, but until they provide pre-built wheels, I don't think we will be able to support this. Even if we did support this, users would have to build statsmodels from scratch, which is very slow. So we'll leave this PR with a hold label until statsmodels supports it, and we will follow that PR closely

AGSaidi commented 2 years ago

Looks like statsmodel has managed to release aarch64 wheels: https://pypi.org/project/statsmodels/#files

@aaronreidsmith @odidev perhaps we can give this another try now?

aaronreidsmith commented 2 years ago

Nice! @odidev if you want to rebase, we can use this branch. Or I have #458 open (which was using a conda-based approach) that we can repurpose

AGSaidi commented 2 years ago

stats model has, as expected, not built Python 3.6 wheels, so in addition to re-basing, we should drop Python 3.6 from the PR and add 3.10.

odidev commented 2 years ago

Done re-basing. Also removed python 3.6 and added python 3.10.

aaronreidsmith commented 2 years ago

Superseded by #458. Pretty much the same, but we changed how we test on aarch64 platforms, since that was the slow step.