This is not an issue, but a simple trick to successfully install SciPy in the same environment the tensorflow_macos in M1.
Should we update this README?
...
Notes
For Macs with M1, the following packages are currently unavailable:
SciPy and dependent packages
Server/Client TensorBoard packages
...
Tricky part
Install all the packages first before the tensorflow_macos script executed
I've been playing around with multiple virtual environments to up and running all the packages I need. So far, it works by managing the packages using miniforge.
# Packages to install via conda (that fails in pip)
# E.g: scikit-learn, scipy
tf_venv="YOUR_VENV"
conda install -c conda-forge -n $tf_venv scipy scikit-learn
Your conda venv:
tensorflow_macos git:(master) ✗ conda info -e
# conda environments:
#
base /opt/homebrew/Caskroom/miniforge/base
YOUR_VENV * /opt/homebrew/Caskroom/miniforge/base/envs/YOUR_VENV
Step to reproduce.
Install Python 3.8 using miniforge via brew or pyenv.
# Install miniforge using brew
brew install miniforge
tf_venv="YOUR_VENV"
Create TF virtual environment using conda
conda create -n $tf_venv python=3.8
Init the shell
conda init "$(basename "${SHELL}")"
Activate the virtual env
conda activate $tf_venv
2. Install the packages using `conda` not `pip`
```shell
# Packages to install via conda (that fails in pip)
# E.g: scikit-learn, scipy
tf_venv="YOUR_VENV"
conda install -c conda-forge -n $tf_venv scipy scikit-learn
Install this repo tensorflow_macos to the existing virtual environment instead of creating the new one.
This line,
Path to new or existing virtual environment [default: /Users/fx/tensorflow_macos_venv/]:
Change to /opt/homebrew/Caskroom/miniforge/base/envs/YOUR_VENV/
Path to new or existing virtual environment [default: /Users/fx/tensorflow_macos_venv/]: /opt/homebrew/Caskroom/miniforge/base/envs/YOUR_VENV/
➜ tensorflow_macos git:(master) ✗ ./scripts/download_and_install.sh
Installation script for pre-release tensorflow_macos 0.1alpha2. Please visit https://github.com/apple/tensorflow_macos
for instructions and license information.
This script will download tensorflow_macos 0.1alpha2 and needed binary dependencies, then install them into a new
or existing Python 3.8 virtual environment.
Continue [y/N]? y
Downloading installer.
/var/folders/22/vz4qt0tn5v59rdq4b653mn000000gn/T/tmp.J6kIvMRX ~/tensorflow_macos
Extracting installer.
Path to new or existing virtual environment [default: /Users/fx/tensorflow_macos_venv/]:
Hi,
This is not an issue, but a simple trick to successfully install SciPy in the same environment the
tensorflow_macos
in M1. Should we update this README?Tricky part
Install all the packages first before the
tensorflow_macos
script executedI've been playing around with multiple virtual environments to up and running all the packages I need. So far, it works by managing the packages using
miniforge
.Your conda venv:
Step to reproduce.
tf_venv="YOUR_VENV"
Create TF virtual environment using conda
conda create -n $tf_venv python=3.8
Init the shell
conda init "$(basename "${SHELL}")"
Activate the virtual env
conda activate $tf_venv
tensorflow_macos
to the existing virtual environment instead of creating the new one. This line,Change to
/opt/homebrew/Caskroom/miniforge/base/envs/YOUR_VENV/
Expected result
and
Thank you.