Docker light images to support Machine Learning (ML) in Python

Overview

That project produces light OCI (Docker-compliant) images, which provide Python environments, ready to use and to develop with Artificial Intelligence (AI), Machine Learning (ML) and Data Science. These images are based on the latest Python-ready Linux distributions.

As well explained in a PythonSpeed article from June 2023, the best light images for Python production operations are based on Debian (the latest stable release being, as of end 2023, Debian 12, also known as Bookworm). For historical reasons (and for the challenge!), Alpine-based distributions are still maintained too.

These Python OCI light images are aimed at deploying Data Science applications on operational environments such as cloud-based Kubernetes clusters or services (e.g., AWS EKS, Azure AKS, IBM/RedHat OpenShift v4 or Google GKE).

The author of this repository also maintains Data Science Python OCI images for every day development purposes in a dedicated GitHub repository and Docker Hub space. Thanks to Docker multi-stage builds, one can easily have in the same Docker specification files two images, one for every day data science work, and the other one to deploy the corresponding applications onto production environments.

The Docker images of this repository just add some standard ML-related Python packages such as NumPy, Pandas and Dash/Plotly, on top of the native images maintained by the Docker Python project:

• Python 3.12
  • Python 3.12 - Debian Bookworm
  • Python 3.12 - Debian Bullseye
  • Python 3.12 - Alpine 3.19
• Python 3.11
  • Python 3.11 - Debian Bookworm
  • Python 3.11 - Debian Bullseye
  • Python 3.11 - Alpine 3.19
• Python 3.10
  • Python 3.10 - Debian Bookworm
  • Python 3.10 - Debian Bullseye
• Python 3.9
  • Python 3.9 - Debian Bookworm
  • Python 3.9 - Debian Bullseye
• Python 3.8
  • Python 3.8 - Debian Bookworm
  • Python 3.8 - Debian Bullseye

In the OCI image, Python packages are installed by the pip utility. For testing purposes, outside of the container, Python virtual environments may be installed thanks to Pyenv and pipenv, as detailed in the dedicated procedure on the Python induction notebook sub-project.

Any additional Python module may be installed either:

• With pip and some requirements.txt dependency specification file:

$ python -mpip install -r requirements.txt

• In a dedicated virtual environment, controlled by pipenv through local Pipfile (and potentially Pipfile.lock) files, which should be versioned:

$ pipenv --rm; pipenv install; pipenv install --dev

On the other hand, the OCI images install those modules globally.

The Docker images of this repository are intended to run any collection of Dash applications.

See also

• Images on Docker Cloud

• Images on Quay.io

• General purpose light Python/Debian/Alpine OCI image: https://github.com/machine-learning-helpers/docker-python-light

• Native Python OCI images:

  • Python 3.12
    • https://github.com/docker-library/python/tree/master/3.12/bookworm
    • https://github.com/docker-library/python/tree/master/3.12/bullseye
    • https://github.com/docker-library/python/tree/master/3.12/alpine3.19
  • Python 3.11
    • https://github.com/docker-library/python/tree/master/3.11/bookworm
    • https://github.com/docker-library/python/tree/master/3.11/bullseye
    • https://github.com/docker-library/python/tree/master/3.11/alpine3.19
  • Python 3.10
    • https://github.com/docker-library/python/tree/master/3.10/bookworm
    • https://github.com/docker-library/python/tree/master/3.10/bullseye
  • Python 3.9
    • https://github.com/docker-library/python/tree/master/3.9/bookworm
    • https://github.com/docker-library/python/tree/master/3.9/bullseye
  • Python 3.8
    • https://github.com/docker-library/python/tree/master/3.8/bookworm
    • https://github.com/docker-library/python/tree/master/3.8/bullseye
  • Python 3.7
    • https://github.com/docker-library/python/tree/master/3.7/bookworm
    • https://github.com/docker-library/python/tree/master/3.7/bullseye

• Debian releases

  • Debian 12 Bookworm
  • Debian 11 Bullseye

• Alpine Linux releases

  • Alpine 3.19
  • Alpine 3.18

Simple use

• Download the Docker image:

$ docker pull infrahelpers/python-light

• Launch Dash or Flask within the Docker image (where <port> corresponds to the local port on which Dash or Flask is launched; the default is 8050):

$ docker run -d -p <port>:8050 infrahelpers/python-light

Interact with the Dash application in a Web browser

The sample Dash application (run from the Docker image) is now available on the Web browser: http://localhost:8050 Note that the port (8050 by default) may be changed as per your convenience.

Build your own container image

• Clone the Git repository:

$ mkdir -p ~/dev/ml && cd ~/dev/ml
$ git clone https://github.com/machine-learning-helpers/docker-python-light.git
$ cd docker-python-light

• Build the OCI images (here with Docker, but any other tool may be used):
  • Python 3.12
    • py312-bookworm tag: Debian Bookworm / Python 3.12:

$ docker build -t infrahelpers/python-light:py312-bookworm python-3.12-bookworm

- `py312-bullseye` tag: Debian Bullseye / Python 3.12:

$ docker build -t infrahelpers/python-light:py312-bullseye python-3.12-bullseye

- `py312-alp319` tag: Alpine 3.19 / Python 3.12:

$ docker build -t infrahelpers/python-light:py312-alp319 python-3.12-alpine-3.19

• Python 3.11
  • py311-bookworm tag: Debian Bookworm / Python 3.11:

$ docker build -t infrahelpers/python-light:py311-bookworm python-3.11-bookworm

- `py311-bullseye` tag: Debian Bullseye / Python 3.11:

$ docker build -t infrahelpers/python-light:py311-bullseye python-3.11-bullseye

- `py311-alp319` tag: Alpine 3.19 / Python 3.11:

$ docker build -t infrahelpers/python-light:py311-alp319 python-3.11-alpine-3.19

• Python 3.10
  • py310-bookworm tag: Debian Bookworm / Python 3.10:

$ docker build -t infrahelpers/python-light:py310-bookworm python-3.10-bookworm

- `py310-bullseye` tag: Debian Bullseye / Python 3.10:

$ docker build -t infrahelpers/python-light:py310-bullseye python-3.10-bullseye

• Python 3.9
  • py39-bookworm tag: Debian Bookworm / Python 3.9:

$ docker build -t infrahelpers/python-light:py39-bookworm python-3.9-bookworm

- `py39-bullseye` tag: Debian Bullseye / Python 3.9:

$ docker build -t infrahelpers/python-light:py39-bullseye python-3.9-bullseye

• Python 3.8
  • py38-bookworm tags: Debian Bookworm / Python 3.8:

$ docker build -t infrahelpers/python-light:py38-bookworm python-3.8-bookworm

- `latest`/`py38-bullseye` tags: Debian Bullseye / Python 3.8:

$ docker build -t infrahelpers/python-light:py38-bullseye python-3.8-bullseye

• (Optional) Push the newly built images to Docker Hub. That step is usually not needed, as the images are automatically built everytime there is a change on GitHub)

$ docker login
$ docker push infrahelpers/python-light:py312-bookworm
$ docker push infrahelpers/python-light:py312-bullseye
$ docker push infrahelpers/python-light:py312-alp319
$ docker push infrahelpers/python-light:py311-bookworm
$ docker push infrahelpers/python-light:py311-bullseye
$ docker push infrahelpers/python-light:py311-alp319
$ docker push infrahelpers/python-light:py310-bookworm
$ docker push infrahelpers/python-light:py310-bullseye
$ docker push infrahelpers/python-light:py39-bookworm
$ docker push infrahelpers/python-light:py39-bullseye
$ docker push infrahelpers/python-light:py38-bookworm
$ docker push infrahelpers/python-light:py38-bullseye
$ docker push infrahelpers/python-light:alp319-py311
$ docker push infrahelpers/python-light:alp318-py311

• (Optional) Push the newly built images to Quay.io. That step is usually not needed, as the images are automatically built everytime there is a change on GitHub)
  • Login to Quay:

$ docker login quay.io

+ Python 3.11:

$ docker tag infrahelpers/python-light:py311-bookworm quay.io/artificialintelligence/python-light:py311-bookworm
$ docker push quay.io/artificialintelligence/python-light:py311-bookworm
#
$ docker tag infrahelpers/python-light:py311-bullseye quay.io/artificialintelligence/python-light:py311-bullseye
$ docker push quay.io/artificialintelligence/python-light:py311-bullseye

+ Python 3.10:

$ docker tag infrahelpers/python-light:py310-bookworm quay.io/artificialintelligence/python-light:py310-bookworm
$ docker push quay.io/artificialintelligence/python-light:py310-bookworm
#
$ docker tag infrahelpers/python-light:py310-bullseye quay.io/artificialintelligence/python-light:py310-bullseye
$ docker push quay.io/artificialintelligence/python-light:py310-bullseye

+ Python 3.9:

$ docker tag infrahelpers/python-light:py39-bookworm quay.io/artificialintelligence/python-light:py39-bookworm
$ docker push quay.io/artificialintelligence/python-light:py39-bookworm
#
$ docker tag infrahelpers/python-light:py39-bullseye quay.io/artificialintelligence/python-light:py39-bullseye
$ docker push quay.io/artificialintelligence/python-light:py39-bullseye

+ Python 3.8:

$ docker tag infrahelpers/python-light:py38-bookworm quay.io/artificialintelligence/python-light:py38-bookworm
$ docker push quay.io/artificialintelligence/python-light:py38-bookworm
#
$ docker tag infrahelpers/python-light:py38-bullseye quay.io/artificialintelligence/python-light:py38-bullseye
$ docker push quay.io/artificialintelligence/python-light:py38-bullseye

• Shutdown the Docker image

$ docker ps
CONTAINER ID IMAGE                    COMMAND                   CREATED        STATUS        PORTS                  NAMES
7b69efc9dc9a ai/python-light          "/bin/sh -c 'python …"    48 seconds ago Up 47 seconds 0.0.0.0:9000->8050/tcp vigilant_merkle
$ docker kill vigilant_merkle
vigilant_merkle
$ docker ps
CONTAINER ID IMAGE                    COMMAND                   CREATED        STATUS        PORTS                  NAMES