This is a docker environment for replicating a standard machine learning dev setup of anaconda + python 3.6 + tensorflow. The official TF docker images are slow to update their Python version, and don't have all the built in stuff you expect in a full featured Anaconda development setup.
The image is a bit hefty for my taste (about 2.5GB at last count) but it gets the job done for running unit tests and doing some basic development. Note that it doesn't utilize the GPU accelerated TF. It installs several specific python libraries that you may not be intersted in, but they're convenient for what I need. The only real trick in here was getting the matching libc++ needed for the latest tensorflow and the opengl libaries needed for running matplotlib.
Check versions by running
docker run --rm -it riansanderson/anaconda-py3-tensorflow python --version && python -c 'import tensorflow as tf; print("TensorFlow v%s"%tf.__version__)'
Play around with an interactive shell that can access your current directory
docker run -it --rm -v ${PWD}:/app riansanderson/anaconda-py3-tensorflow /bin/bash
If you use matplotlib to generate plots to files you'll need to configure matplotlib to use a non-interactive backend. There are several ways to do it, I put this single line in a matplotlibrc file in the unittest directory where tests were being run out of:
backend : agg