Custom TensorFlow Docker setup

Building the docker image on top of tensorflow1.6-gpu-py3

Base image of official TensorFlow 1.6(GPU, Py3) is used. Please visit https://hub.docker.com/r/tensorflow/tensorflow/tags/ for official TensorFlow images.

Steps to create custom docker image:

1. Create a directory and copy Dockerfile. Dockerfile defines what goes on in the environment inside your container.
2. Dockerfile starts with base image, in this case FROM gcr.io/tensorflow/tensorflow:1.6.0-gpu-py3.
3. Additional dependencies can be included on top of base image e.g. RUN apt-get update && apt-get install -y --no-install-recommends python3-tk. RUN command is used to execute installation. If dependencies are stored in additional file, we have to first copy it (e.g. requirements.txt) to docker container with the help of COPY command.
4. Likewise pip or anyother package management system can be used to install dependencies. e.g. RUN pip install --trusted-host pypi.python.org -r requirements.txt . In this case dependencies are written in requirements.txt file.
5. Before loading the container, other initial settings can be applied e.g. WORKDIR / changes the current working directory to root.
6. For building docker, using terminal go to directory where Dockerfile is stored. Now run the build command i.e. docker build -t friendlyhello ., this creates a Docker image. This image gets stored in local Docker image registry.