TensorFlow

When submitting a job to the grid (which is the future goal), it will initialise a virtual Docker container which the code will run in. To develop in, this isn't ideal.

The first part shows you how to install TensorFlow locally without Docker so that developing code is more easy (don't have to copy to and from Docker containers, can host code in Git etc.).

The second part demonstrates installing TensorFlow in a Docker container on your mac, with the intention of mimicking what the grid environment will contain.

TensorFlow for developing on Mac

• sudo easy_install pip
• sudo pip install --upgrade virtualenv

• virtualenv --system-site-packages ~/tensorflow

• source ~/tensorflow/bin/activate

• pip install --upgrade https://storage.googleapis.com/tensorflow/mac/tensorflow-0.7.1-cp27-none-any.whl

• deactivate

TensorFlow packages and examples are housed in ~tensorflow/lib/python2.7/site-packages/tensorflow/models/image/cifar10. Whenever you want to use TensorFlow you have to source your env: source ~/tensorflow/bin/activate.

TensorFlow mimicking the Grid env

Docker

Docker has a good instruction page here: https://docs.docker.com/engine/installation/mac/ But essentially:

• Go here: https://www.docker.com/products/docker-toolbox and click Download, then follow instructions.

Hello from Docker.
This message shows that your installation appears to be working correctly.

To generate this message, Docker took the following steps:
 1. The Docker client contacted the Docker daemon.
 2. The Docker daemon pulled the "hello-world" image from the Docker Hub.
 3. The Docker daemon created a new container from that image which runs the
    executable that produces the output you are currently reading.
 4. The Docker daemon streamed that output to the Docker client, which sent it
    to your terminal.

To try something more ambitious, you can run an Ubuntu container with:
 $ docker run -it ubuntu bash

Share images, automate workflows, and more with a free Docker Hub account:
 https://hub.docker.com

For more examples and ideas, visit:
 https://docs.docker.com/userguide/

When starting a new Docker image, one can provide interactive flags (-i). To detach from this shell without it exiting, a Ctrl-d will do the trick (docker info) will tell you how many containers are running. To list all containers (stopped and running) do docker ps -a.

Some more interactive commands can be found here: https://docs.docker.com/engine/quickstart/#running-an-interactive-shell

Port forwarding

https://localhost:8010

So Docker is now installed, now for TensorFlow

TensorFlow

Installation of TensorFlow in the Docker environment is incredibly easy. Type docker run -it b.gcr.io/tensorflow/tensorflow:latest-devel from the "docker Quickstart Terminal". On first run, this will download the TensorFlow image, subsequent runs use the same, fresh image. More info and a list of images can be found at https://www.tensorflow.org/versions/r0.7/get_started/os_setup.html.

To get actual tag names that you can download do:

• wget https://gcr.io/v2/tensorflow/tensorflow/tags/list

To test if TensorFlow properly installed do python -m tensorflow.models.image.mnist.convolutional. Output should look something like:

Successfully downloaded train-images-idx3-ubyte.gz 9912422 bytes.
Successfully downloaded train-labels-idx1-ubyte.gz 28881 bytes.
Successfully downloaded t10k-images-idx3-ubyte.gz 1648877 bytes.
Successfully downloaded t10k-labels-idx1-ubyte.gz 4542 bytes.
Extracting data/train-images-idx3-ubyte.gz
Extracting data/train-labels-idx1-ubyte.gz
Extracting data/t10k-images-idx3-ubyte.gz
Extracting data/t10k-labels-idx1-ubyte.gz
Initialized!
Step 0 (epoch 0.00), 7.4 ms
Minibatch loss: 12.053, learning rate: 0.010000
Minibatch error: 90.6%
Validation error: 84.6%
Step 100 (epoch 0.12), 601.3 ms
...

A nice image recognition example for TensorFlow can be found here: https://www.tensorflow.org/versions/r0.7/tutorials/image_recognition/index.html

More examples/tutorials can be found here: https://www.tensorflow.org/versions/r0.7/tutorials/index.html

Tensorboard

• tensorboard --logdir ./ --port 8888

http://localhost:8810/

-- JoshuaWyattSmith - 2016-04-01