pip install does not work on Mac

[erwtokritos]

Hey there,

First of all, congratulations for your effort, this is a great initiative!

I am raising this issue because I have faced a problem with installation. I have created a Python 3.8.6 virtual environment on my Mac and installed tensorflow 2.5.0 successfully. When I ran the installation command for the "Tensorflow Decision Forests" package, pip3 install tensorflow_decision_forests --upgrade

I got:

ERROR: Could not find a version that satisfies the requirement tensorflow_decision_forests (from versions: none) ERROR: No matching distribution found for tensorflow_decision_forests

It's a bit confusing because the installation command on PyPi (I guess this is the right one) contains dashes ,instead of underscores, in the package name.

Any ideas?

Thanks a lot

[informatica92]

Same stuff on Windows too. Let me add some other details of the environment where I installed tensorflow 2:

• python version: 3.8.8
• tensorflow version: 2.5.0
• pip version: 21.1.2

I also noticed the same confusion between dashes and underscores. Seems strange to me

[informatica92]

Looking at the files available in pip (here: https://pypi.org/project/tensorflow-decision-forests/#files) I noticed that all the available builds, have been made for linux only (python 3.6, 3.7, 3.8 and 3.9). So basically, no Windows and Mac support up to now

[janpfeifer]

hi Thanos,

Indeed the Mac version is in our TODO list, as is with windows (see issue #3 ), hence pip not finding a matching tensorflow_decision_forests version.

We'll keep this issue opened and use it to post updates. Sorry about that.

[erwtokritos]

Thank you for your prompt response Looking forward to start playing with the library :)

Have a good day!

[dkgaraujo]

For Mac/Windows users: a quick fix to get your feet wet while a version in pip is being prepared is to use the model in Google Colab, or if you are an advanced user that really wants to use a local install, you can create a docker with ubuntu.

(Incidentally, @janpfeifer no need to apologise! Best to put a version out there and keep adding more things than wait. Thanks very much to you and your team for your work on tfdf)

[ericzhangcn1980]

@janpfeifer Look forward to the MAC version update, thank you!

[AlirezaSadeghi]

Hey @janpfeifer, are there any updates or rough timelines you might have on this? As now TFX also has experimental support for tfdf, I think a mac runnable version is very much in need to facilitate a good dev experience.

[janpfeifer]

hi @AlirezaSadeghi, this was very timely, since @achoum just released the Mac version yesterday, v0.2.3:

https://github.com/tensorflow/decision-forests/releases/tag/0.2.3

Let us know how it goes -- it's new, so there may be some initial bumps (hopefully none). And apologies it took so long, there were some issues in TF build process that only now got sorted out (afaik).

[AlirezaSadeghi]

Super. Thanks @janpfeifer and will do!

[janpfeifer]

The Mac version should be in pip:

https://pypi.org/project/tensorflow-decision-forests/#files

[JanetVictorious]

I'm still unsuccessful in installing from pip:

Python 3.8.10
pip 22.0.3

When I try to install from pip (pip install tensorflow-decision-forests==0.2.3) I get the following error:

ERROR: Could not find a version that satisfies the requirement tensorflow-decision-forests==0.2.3 (from versions: none)
ERROR: No matching distribution found for tensorflow-decision-forests==0.2.3

Is this expected?

[achoum]

Currently, TF-DF is only exported for python 3.9 on MacOS (see https://pypi.org/project/tensorflow-decision-forests/#files for the list of supported versions).

Until this is extended to other python version, can you try installing it again, but with python 3.9?

[cacacoo]

Hi. I was going to get the whl file and install it (https://pypi.org/project/tensorflow-decision-forests/#files)

but i got: ERROR: tensorflow_decision_forests-0.2.3-cp39-cp39-macosx_12_0_x86_64.whl is not a supported wheel on this platform. also ERROR: tensorflow_decision_forests-0.2.4-cp39-cp39-macosx_12_0_x86_64.whl is not a supported wheel on this platform.

My environment:

macos big sur 11.6
python version: 3.9.10 with anaconda
tensorflow version: 2.7.0
pip version: 22.0.3

I think I got the right whl file, but it's weird. Does anyone know the reason?

[JanetVictorious]

Thank you @achoum, that did the trick!

[achoum]

@cacacoo

My best guess is the version of MacOS (macosx 12 vs macosx 11). I am not certain there is effectively a difference (like for the version of python), and I would try renaming the .whl file and editing /tensorflow_decision_forests-0.2.4.dist-info/WHEEL accordingly.

Ultimately, we will provide all the right versions directly in pip :), so thanks for your patience with all of this.

[b3rge]

Hi!

I am running: pip3 install tensorflow_decision_forests-0.2.4-cp39-cp39-macosx_12_0_x86_64.whl

on macos 12.2 Monterey with Python 3.9.0 on a 64bit architecture.

but I still get: ERROR: tensorflow_decision_forests-0.2.4-cp39-cp39-macosx_12_2_x86_64.whl is not a supported wheel on this platform.

Anyone else?

[ericzhangcn1980]

me too!

[cacacoo]

@achoum Thank you!

I modified the whl file name from 12 to 11 to fit my macos version and pip installed it. tensorflow_decision_forests-0.2.4-cp39-cp39-macosx_11_0_x86_64.whl

editing /tensorflow_decision_forests-0.2.4.dist-info/WHEEL accordingly. Can you explain this in more detail? Is there a problem if I don't do it?

[ynait]

Upvoting a mac version for py3.8

[achoum]

Hi,

Python 2.8 @cacacoo

TF-DF is now released for python 2.{7,8,9} on MacOS12

See the files: https://pypi.org/project/tensorflow-decision-forests/#files

@b3rge @ericzhangcn1980

Unfortunately, we are currently lacking MacOS experience, therefore we don't have a good solution to those issues. If you figure them out and share the results though, this would be awesome.

My 2 cents:

According the the exported Pip file for macos (see https://pypi.org/project/tensorflow-decision-forests/#files), we currently support :"macosx_12_0_x86_64" for python 2.{7,8,9}.

@b3rge

In your error, the macos version is macosx_12_2_x86_64. I am not sure of what the "2" means (maybe a new sub version), but it would be strange not to have backward compatibility between these two versions. My best guess is that we/you need to find a way to force your pip to install this version.

In addition to the filename, the only other version specific information is in the file "/tensorflow_decision_forests-0.2.4.dist-info/WHEEL". I would not try this approach first, but editing the filename the WHEEL file content might work.

[omit-ai]

Hi all,

since we're working with macOS I was thrilled to get the support! We may have found a workaround, for the time being, for this wheel files.

Setup

1. macOS Monterey Version 12.2.1

• Processor: 2,6 GHz 6-Core Intel Core i7

2. Python Version: 3.9.5
3. pip Version: 22.0.3

Workaround

1. Download tensorflow_decision_forests-0.2.4-cp39-cp39-macosx_12_0_x86_64.whl from here
2. Rename file to tensorflow_decision_forests-0.2.4-cp39-none-any.whl
3. $ pip install tensorflow_decision_forests-0.2.4-cp39-none-any.whl

After that I was able to import tfdf and train a model.

Best regards,

Timo

[janpfeifer]

Thanks @Timo3as for posting your workaround here. Indeed we lack MacOS expertise in our small team :\

Let's keep the issue opened until we find a way to test and release pip packages to the various MacOS versions.

[tlapusan]

Thanks @Timo3as, your workaround worked for me also.

[varunkarwa]

@janpfeifer, can you tell me when tfdf will be available for Windows?

[janpfeifer]

hi @varunkarwa , we are actually waiting for TensorFlow Custom Ops to work again for Windows -- TF-DF uses a "custom op" for Tensorflow.

But at home I have a windows box and I use TF-DF always on WSL (windows sub-system for linux), where I run the kernel for my notebooks. It works really nicely, I recommend. And in WSL2 it must be almost as fast I would think. (Generally my models at home are smaller, so it's always very fast).

#3 is the issue tracking WIndows support.

Also notice that Yggdrasil, the underlying C++ implementation of the algorithms, work for WIndows. So you can link it to do inference in a windows program for instance.

[pimatrix]

me too，not work on Mac OS m1

[tlapusan]

me too，not work on Mac OS m1

check comment from @Timo3as. His workaround worked for me on Mac.

[pimatrix]

me too，not work on Mac OS m1

check comment from @Timo3as. His workaround worked for me on Mac.

Yes it works on Intel Core i7，but failed on Mac of m1

[sun1638650145]

Has anything changed now?

[janpfeifer]

hi all, just wanted to update on unfortunately "no updates" for M1 yet. TensorFlow on M1 doesn't make it easy to support "TensorFlow Custom Ops", which we need for TF-DF :( ... since we don't yet have an update on this on TF side, we can't provide one for TF-DF either.

A question for those interested: if for Mac M1 we had a Decision Forests model working in pure Python -- mostly with the same API, just not integrated into TensorFlow.

The model could be used with numpy, matplotlib, notebook etc. And after training/developing with these models, it could be exported for TF Serving (for Linux). Would such a stop-gap solution be useful ?