tensorflow/decision-forests

Best practice for installation via poetry using M1 mac (arm chip)

jamesowers-roo opened this issue · 2 comments

I've managed to get the installation via poetry to work by explicitly pointing to the wheel URL e.g.

[tool.poetry.dependencies]
tensorflow-decision-forests = [
    { markers = "sys_platform != 'darwin' or platform_machine != 'arm64'", version = "==1.3.0" },
    { markers = "sys_platform == 'darwin' and platform_machine == 'arm64'", url = "https://files.pythonhosted.org/packages/7a/a4/e818814fec37812222f77ae85b41bd6a95c4c90b0b92e03c9c9577601056/tensorflow_decision_forests-1.3.0-cp310-cp310-macosx_12_0_arm64.whl" }
]

The reason the installation via a simple

[tool.poetry.dependencies]
tensorflow-decision-forests = "==1.3.0"

doesn't work, is that tensorflow-decision-forests expects the dependency tensorflow (which doesn't exist for arm mac, one needs tensorflow-macos instead). This dependency is specified correctly within the mac arm64 wheel distributed here: https://pypi.org/project/tensorflow-decision-forests/#files

Is there a better way to achieve this? e.g. is there a documented installation option e.g. pip install 'tensorflow-decision-forests==1.3.0[macos] which I have missed?

rstz commented

Hi,
I'm not very familiar with Poetry, so I believe that this is the best strategy that's available at this point.

If anyone can open a PR that opens a better way - maybe some changes to our setup.py - I'll be happy to integrate it into our next release!

rstz commented

Since tensorflow seems to have consolidated into one package (no need for tensorflow-macos anymore), I expect this to be fixed with the next version.