Galaxy

Galaxy is a provisioning environment for various machine/deep learning software and is intended mainly for local development/experimentation. Currently, the following are available for set-up:

• Berkeley Caffe
• Neural Style
• OpenCV 3.1.0
• Style Transfer
• Deep Visualization Toolbox

Caveats

This tool is for Mac OSX only at the moment. While familiarity with Ansible and Vagrant are not mandatory for set-up, any experience with these provisioning tools will be helpful when attempting to edit/improve/debug the code. One final important note - currently the provisioning steps for any of the above listed software default to a CPU-only installation, as it is assumed that the set-up will be done on a personal machine. A command-line option to switch to a GPU-based installation will be added eventually for those with CUDA enabled graphics cards.

Prerequisites

Package Manager

• Homebrew
• Homebrew Cask

Installation of homebrew and cask will not be covered but can be found here and here.

Python/pip

If you don't already have python and pip installed:

brew install python

Provisioning Tools

• Vagrant - For local, portable development environments
• Virtualbox - Free VM software that acts as a provider for Vagrant (VMWare Fusion is another popular option but requires a license AND a plug-in which also costs money)
• Ansible - For automating software installation/configuration

Install vagrant:

brew cask install vagrant

Install virtualbox:

brew cask install virtualbox

Install ansible:

pip install ansible

Getting Started

While there isn't necessarily a right or wrong way to organize an Ansible project, there are best practices, but for the purposes of this example we will divide the project into apps and roles. Roles are essentially individual pieces of software while apps are environments which are made up of any number of roles.

We have one example app that contains two items:

• a share folder which is optional, but can be used to share files between the host and the guest (the vm) if necessary
• a Vagrantfile, used to configure your Vagrant instance

There are a few important things to note in the Vagrantfile:

• The ip setting should be set to an available host IP address and should match what you see in the ansible/inventory/hosts file
• The cpus and memory settings should be set as needed, but I believe Vagrant caps these values at 50% of your host's capabilities as a safety measure
• The ansible.playbook setting points to the location of the app playbook and within that file you may select the roles you'd like to install (the example only installs style transfer)

Boot up the VM

From within the specific apps directory, so for this example apps/example, run the following:

vagrant up

The first time you do this be prepared to wait as the base ubuntu image needs to be downloaded which is ~1.2GB.

After the image has been downloaded and the VM has been booted up, Vagrant will automatically run the playbook listed in the Vagrantfile. The first time you do this can be time consuming so just sit back and grab a beer.

If there are no issues, you will now be able to login to the VM:

vagrant ssh

Voilà! Navigate to the folder where Style Transfer has been installed /opt/style_transfer (you can configure this location yourself from within the role) and have fun.

License

Galaxy is MIT Licensed.