A Docker image for ARM devices with Tensorflow 1.6.0 an open source software library for numerical computation using data flow graphs that will let you play and learn distinct Machine Learning techniques over Jupyter Notebook an open-source web application that allows you to create and share documents that contain live code, equations, visualizations and explanatory text. Computational Narratives as the Engine of Collaborative Data Science. All this under Python 3.4 language. There image is a copy on another image based on Python 2.7 elswork/rpi-tensorflow You will find there extended information about this image.

Be aware! This image is experimental, you should use the other one unless the use of Python 3 would be indispensable!

You will see some warning messages when executing commands in jupyter notebooks this is caused because tensorflow wheel package is built for Python 3.4 and the Python version of container is 3.5, but despite this commands will work as expected.

Most tags of this container should work under arm32v7 (raspberry, odroid, ...), amd64 tag is for x86 (most computers)

• GitHub
• Docker Hub
• Deft.Work my personal blog

Build for arm32v7 architecture

docker build -t elswork/rpi-tensorflow-py3:latest .

Build for amd64 architecture

docker build -t elswork/rpi-tensorflow-py3:latest \
 --build-arg WHL_URL=https://storage.googleapis.com/tensorflow/linux/cpu/ \
 --build-arg WHL_FILE=tensorflow-1.6.0-cp34-cp34m-linux_x86_64.whl .

In order everyone could take full advantages of the usage of this docker container, I'll describe my own real usage setup. For amd64 architecture replace latest by amd64 tag.

docker run -d -p 8888:8888 elswork/rpi-tensorflow-py3:latest

A more complex sample:

docker run -d -p 8888:8888 -p 0.0.0.0:6006:6006 \
 --restart=unless-stopped elswork/rpi-tensorflow-py3:latest

Point your browser to http://localhost:8888

First time you open it, you should provide a Token to log on you cand find it with this command:

docker logs container_name

With the second example you can run TensorBoard executing this command in the container:

tensorboard --logdir=path/to/log-directory --host=0.0.0.0

And pointing your browser to http://localhost:6006