Fix for builds on Mac M1 infrastructure #641
Conversation
|
|
|
The change in _builder.pyx also solves the cython compiling error on windows. Thanks for the pr, hope it gets merged. |
|
ah fiddle dee dee, let me see what broke |
…troubleshooting) * numpy 1.21.4 - Fix for arm64 Darwin install (Mac M1) * scipy 1.10.0 - Updated numpy dependency * scikit-learn - 1.0.2 Updated numpy dependency * stasmodels 0.13.1 - No install if numpy existing (moving above to conda install) * torch 1.8.0 - Unavailable on python 3.9
| @@ -5,6 +5,7 @@ dependencies: | |||
| - pip | |||
| - python=3.9 | |||
| - setuptools=52.0.0 | |||
| - numpy=1.22 | |||
There was a problem hiding this comment.
Curious to know, what is the difference between putting numpy in this section, vs. the pip section? You don't want to specify a minor version like for all the other dependencies?
There was a problem hiding this comment.
The dependencies listed above will be installed using conda, while the ones in the pip section will use pip. If available, conda installation is more robust as it packages binaries and executables necessary for installation together with the Python library, but not all packages are available in conda.
There was a problem hiding this comment.
Hey @eugeneyarovoi , so initially I'd added above to conda install since numpy was required for pip wheel building prior to installs and saw failures for anything before numpy==1.22.0. I tested again recently and after updating all of the subsequent package versions and looks like it's fine, but if there's a preference for conda as the package handler re: @jeongyoonlee can leave as is. Did find a few other items breaking that I'd like to update quick for the arm64x build on mac so might have a quick fast-follow
* Fix for build breaks (see uber#534) * Removing version from pip (caused errors in `conda env create`) * Myraid of fixes for `conda env create` method (defaulted to this for troubleshooting) * numpy 1.21.4 - Fix for arm64 Darwin install (Mac M1) * scipy 1.10.0 - Updated numpy dependency * scikit-learn - 1.0.2 Updated numpy dependency * stasmodels 0.13.1 - No install if numpy existing (moving above to conda install) * torch 1.8.0 - Unavailable on python 3.9
Proposed changes
Following up with issues I saw with pip install in M1 architecture. Deep dived the issue trying
conda env createand found issues there as well, so added some fixes for that build on M1 (mostly updating for compatability withnumpyfor Darwin arm64 sys build). Breaking out as separate commits to simplify cherry picking since some issues may not be generalizable. Known related issue #534.Types of changes
What types of changes does your code introduce to CausalML?
Put an
xin the boxes that applyChecklist
Put an
xin the boxes that apply. You can also fill these out after creating the PR. If you're unsure about any of them, don't hesitate to ask. We're here to help! This is simply a reminder of what we are going to look for before merging your code.Further comments
None, other than the fact that this is my first PR and I would not be surprised if there's a more elegant resolution