Setting Up a Python environment on Mac OS for AI/ML Development

implementation
Imported and normalized from a Notion article.
Author

MUHAMMAD GHIFARY

Published

January 26, 2025

I’ve lost count of how many times I’ve set up Python environments on macOS, yet I still sometimes struggle to remember all the steps to recreate the suitable setup, especially when starting fresh on a new machine. So, I decided to jot down the process—both for myself and for anyone else looking to configure their Python environment, particularly for AI and machine learning projects. Let’s dive in!

  1. Installing Homebrew

Homebrew is a fantastic package manager that simplifies software program installation via the command line on macOS (and Linux, too). Think of it as your app store, but for command-line tools. I follow the guideline from https://mac.install.guide/homebrew/3.

To get started, open Terminal and run:

bash$/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

Once installed, we integrate Homebrew into $PATH by executing these two commands:

bash$(echo; echo 'eval "$(/opt/homebrew/bin/brew shellenv)"') >> ~/.zprofile$eval "$(/opt/homebrew/bin/brew shellenv)"

To ensure everything is set up correctly, check Homebrew’s status:

bash$brew doctor

Now we can see

Your system is ready to brew.
  1. Installing Python with Pyenv

While Homebrew allows for direct Python installation (can easily done through brew install python), managing multiple Python versions across various projects can be cumbersome. Homebrew-installed Python is not well-suited to perform such tasks.

I prefer using another tool like Pyenv. It’s a tool designed specifically for managing multiple Python versions, which is a lifesaver when juggling projects that require different versions of Python.

Here’s how I set it up.

First, install Pyenv with Homebrew:

bash$brew install pyenv

Next, configure our shell to initialize Pyenv. Add the following lines to ~/.zprofile

export PYENV_ROOT="$HOME/.pyenv"
[[ -d$PYENV_ROOT/bin ]] && export PATH="$PYENV_ROOT/bin:$PATH"
eval "$(pyenv init -)"

After saving the file, reload the shell configuration source ~/.zprofile or just restart the terminal.

Check if the Pyenv is properly installed.

$pyenv --version pyenv 2.5.1

Before using pyenv to install Python, we must use Homebrew to install the xz package. Pyenv’s Homebrew formula does not specify all needed dependencies during installation with Homebrew.

$brew install xz

Now, we can comfortably install any version of Python with pyenv. For example,

$pyenv install 3.11

We can see a list of installed Python versions.

$pyenv versions

Interestingly, I still get the following error when executing which python or python --version.

python not found

It turns out that I just need to setup some other environment variables related to Pyenv. So no need to be panic!

In doing so, open ~/.zshrc and ensure it contains:

export PYENV_ROOT="$HOME/.pyenv"
export PATH="$PYENV_ROOT/bin:$PATH"
eval "$(pyenv init --path)"
eval "$(pyenv init -)"

Then save the file and reload the shell source ~/.zshrc or restart the terminal.

Now I can see something like this:

$which python /Users/mghifary/.pyenv/shims/python$python --version Python 3.11.11

Success! Pyenv is now managing our Python installation.

  1. Setting Up a Virtual Environment

Creating a virtual environment ensures that our project dependencies are isolated, preventing conflicts. Here’s how:

Create a virtual environment by executing the venv module:

python -m venv ~/.tf-metal

Activate the environment:

source ~/.tf-metal/bin/activate

The terminal prompt should now reflect the activated environment, typically by prefixing it with (.tf-metal).

  1. Installing TensorFlow with GPU Support on macOS

Once the virtual environment is ready, we can install some relevant libraries for implementing and running AI/ML programs. To leverage GPU acceleration on macOS, especially with Apple Silicon, we’ll need to install TensorFlow along with the Metal plugin. Here’s the process:

Install TensorFlow for macOS:

(.tf-metal)$python -m install tensorflow-macos

Install the TensorFlow Metal plugin:

(.tf-metal)$python -m install tensorflow-metal

These installations enable TensorFlow to utilize the GPU via Apple’s Metal API.

It’s essential to confirm that TensorFlow can detect our GPU. Here’s a quick test:

import tensorflow as tf
print(tf.__version__)

# List physical devices
physical_devices = tf.config.list_physical_devices('GPU')
print("Num GPUs Available: ", len(physical_devices))

# Check if Metal is being used
for device in physical_devices:
    print(device)
2.16.2
Num GPUs Available:  1
PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')

Note: As of now, it seems that TensorFlow with Metal support is compatible up to Python 3.11. Attempts to install tensorflow-metal with Python 3.12 were unsuccessful.

Now we’ve established a robust Python environment on the macOS system, optimized for AI/ML development with GPU acceleration.

Happy coding! 💻