A docker-based starter kit for machine learning via jupyter notebooks

Docker Images

To support both old and new environments, docker images covers various combinations of

• machine learning frameworks (Keras, Tensorflow, PyTorch)
• CUDA v8, v9 and v10.

Python 3 only as Python 2 is end-of-life, so deprecated.

All of images include vision-centric libraries, such as

• JupyterLab
• OpenCV
• scikit-image

Check the CUDA compatibility chart for the required version of Nvidia garphics driver for your host system.

Tags

If you are reading this page from Docker Hub, the links to Dockefiles will not work. Please go to the github project page instead.

Experimental

Tag	Comment	Dockerfile	Info
latest	CPU-only. Jupyter. PyTorch 1.4, Keras, TF 1.15.2.	Dockerfile
latest-gpu	CUDA 10.1. Jupyter. PyTorch 1.4, Keras, TF 1.15.2.	Dockerfile

Caffe

PyTorch

Images of Pytorch version 1.5 and higher include Pytorch Lightning.

Tag (OS-based python)	Comment	Dockerfile	Info
pytorch1.6	CPU-only	Dockerfile
pytorch1.6-cuda101	Nvidia Driver >= 418.xx	Dockerfile
pytorch1.6-cuda92	Nvidia Driver >= 396.xx	Dockerfile
jupyter-pytorch1.2-py3-cuda10	Nvidia Driver >= 410.xx	Dockerfile
jupyter-pytorch1.1-py3-cuda9	Nvidia Driver >= 384.xx	Dockerfile
jupyter-pytorch1.0-py3-cuda8	Nvidia Driver >= 375.xx	Dockerfile

Tag (Conda-based python)	Comment	Dockerfile	Info
jupyter-pytorch1.3-conda3	CPU-only	Dockerfile
jupyter-pytorch1.3-conda3-cuda92	Nvidia Driver >= 396.37	Dockerfile
jupyter-pytorch1.1-conda3-cuda9	Nvidia Driver >= 384.xx	Dockerfile
jupyter-pytorch1.0-conda3-cuda8	Nvidia Driver >= 375.xx	Dockerfile

Tensorflow (including Keras)

Tag (OS-based python)	Comment	Dockerfile	Info
tf2.3	CPU-only	Dockerfile
tf2.3-cuda101	Nvidia Driver >= 418.xx	Dockerfile
tf2.0.0-cuda10	Nvidia Driver >= 410.xx	Dockerfile
tf1.15.2	CPU-only	Dockerfile
tf1.15.2-cuda10	Nvidia Driver >= 410.xx	Dockerfile
jupyter-keras-tf1.12.3-py3-cuda9	Nvidia Driver >= 384.xx	Dockerfile
jupyter-keras-tf1.4.1-py3-cuda8	Nvidia Driver >= 375.xx	Dockerfile

Tag (Conda-based python)	Comment	Dockerfile	Info
jupyter-keras-tf1.14.0-conda3	CPU-only	Dockerfile
jupyter-keras-tf1.14.0-conda3-cuda10	Nvidia Driver >= 410.xx	Dockerfile
jupyter-keras-tf1.12.0-conda3-cuda9	Nvidia Driver >= 384.xx	Dockerfile
jupyter-keras-tf1.4.1-conda3-cuda8	Nvidia Driver >= 375.xx	Dockerfile

Internal Tags

For intermediate Docker images, from which final images are build from, see INTERNAL.md.

Deprecated Tags

For older versions, see deprecated/deprecated.md.

Usage

Step 1: pull pre-built images:

docker pull wqael/notebooks:<tag>

Step 2: launch image:

docker run -it -v $2:/notebooks -p 8888:8888 -p 6006:6006 $1

or, for GPU support

nvidia-docker run -it -v $2:/notebooks -p 8888:8888 -p 6006:6006 $1

where:

• $1 is the tag for a docker image, e.g. wqael/notebooks:latest.
• $2 is the folder containing the notebooks on the host file system, e.g. clone this repo and use ~/notebooks.

Step 3: From the log, copy-and-paste the line similar to the following to your favorite browser:

    Copy/paste this URL into your browser when you connect for the first time,
    to login with a token:
        http://localhost:8888/?token=<token string>

Bonus step: Use next generation Jupyter:

After jupyter home page is loaded, i.e. http://localhost:8888/tree, browse to http://localhost:8888/lab.

Step 4: How to shutdown the docker image:

In the running image terminal, hit Ctrl+C twice.