Using Tensorflow with Docker

I started the Google Deep Learning Course featuring the Python Tensorflow library.

Now I’m using a Windows 7 laptop, and the installation instructions for Windows on the Tensorflow website are a bit less developed than for unix based systems. The Deep Learning course comes with a Docker image packing tensorflow and the course exercises in an IPython notebook, and this seems to actually be the simplest way to get Tensorflow on Windows.

To install and get started:

  1. Download and install the Docker Toolbox
  2. Open the Docker Quickstart terminal
  3. Paste the following command in this terminal. This will download (or open if it is already downloaded) the docker image containing Tensorflow and the course exercises.
    docker run -p 8888:8888 -it --rm b.gcr.io/tensorflow-udacity/assignments:0.4.0
  4. At the top of your Docker terminal you should see an IP address: go to http://<this_ip_adress>:8888 in your browser. This should be the home page of the Jupyter notebook featuring the course exercises. You will be able to import tensorflow when coding through this interface.

I tried this install on a mac and it works too !