ericjang/TensorFlow_Windows.md

## TensorFlow_Windows.md

      
    Raw
  

              TensorFlow_Windows.md
            
          
    TensorFlow development environment on Windows using Docker

Here are instructions to set up TensorFlow dev environment on Docker if you are running Windows, and configure it so that you can access Jupyter Notebook from within the VM + edit files in your text editor of choice on your Windows machine.
Installation

First, install
https://www.docker.com/docker-toolbox
Since this is Windows, creating the Docker group "docker" is not necessary.
Launching TensorFlow

Open up "Docker Quickstart Terminal". It should bring up a MinGW-type shell.
You could also use Windows cmd.exe or Powershell.exe but they haveadditional configurations that you need to do before you can run docker.exe
Launch the TensorFlow container.
Set up port forwarding, mount contents of "thesis" directory into /home (the thesis folder doesn't show up, as its contents are mapped to the contents of /home.
$ docker run -it -p 8888:8888 -p 6006:6006 -v //c/Users/eric/Dropbox/thesis:/home b.gcr.io/tensorflow/tensorflow

The first time the command is run, it will Download and install TensorFlow. Afterwards,
This should bring you into a Linux VM.
The command above mounts a folder in your Windows host machine into the container. It's preferable to do things this way, because the container does not persist your files by default.
It already comes with Jupyter installed, but you won't be able to access the notebooks by navigating
to localhost:8888 (or whatever port Jupyter starts on). You will also want to expose port 6006 to be able to use TensorBoard (currently, exposing a port on a live container is not possible).
Instead, since it is running in a VM, not only do you need to forward the port (hence the -p 8888:8888), but the IP address you access
needs to be the IP address of the VM, not the Windows Machine. Hence, you need to find the address of the docker machine running the container.
Listing the docker containers
$ docker ps
CONTAINER ID        IMAGE                            COMMAND             CREATED             STATUS              PORTS                              NAMES
4dd413d71153        b.gcr.io/tensorflow/tensorflow   "/bin/bash"         2 minutes ago       Up 2 minutes        6006/tcp, 0.0.0.0:8888->8888/tcp   lonely_engelbart

$ docker ps

docker-machine ls
$ docker-machine ls
NAME      ACTIVE   DRIVER       STATE     URL                         SWARM   ERRORS
default   *        virtualbox   Running   tcp://192.168.99.100:2376
$ docker-machine ip default
192.168.99.100

Navigate web browser to 192.168.99.100:8888 (or whatever webserver port your web app is running on) and you should be able to see your web apps.
Hacking on Source Code

Let's say you want to use this container for your research environment. You'll want to clone the docker container so you can make changes to it.
In the Docker Quickstart terminal,
docker pull b.gcr.io/tensorflow/tensorflow-full

$ docker images
REPOSITORY                            TAG                 IMAGE ID            CREATED             VIRTUAL SIZE
b.gcr.io/tensorflow/tensorflow-full   latest              edc3d721078b        3 weeks ago         2.284 GB
b.gcr.io/tensorflow/tensorflow        latest              217daf2537d2        4 weeks ago         652.6 MB
hello-world                           latest              0a6ba66e537a        7 weeks ago         960 B

Let's say you installed a python package like ipdb and want to commit changes. Just do:
docker commit lonely_engelbart ejang/tensorflow

Subsequently,
docker run ejang/tensorflow

Moar Terminals

If you want to attach another shell to the docker container:
docker exec -it lonely_engelbart bash

Increasing Memory on Docker Machine

The docker-VM provided has default 1G memory, which is not sufficient to run the MNIST/CNN examples. You can configure the resource limits of the default Docker-machine VM from Virtualbox.
See this page for more tips:
http://blog.pavelsklenar.com/5-useful-docker-tip-and-tricks-on-windows/