Universe on Ubuntu Chef Cookbook

v0.2.0 (In Development)

The Universe on Ubuntu cookbook creates a virtual machine with Ubuntu 16.04LTS or 14.04LTS as operating system and configures all environment needed to make it possible to use OpenAI Universe as a software platform to measure and train Artificial Intelligence on an extremely wide range of real-time, complex and heterogeneous environments.

More info about Universe: https://openai.com/blog/universe/

Attributes

default['universe']['gpu'] = false

The GPU attribute is set to false by default enabling only CPU processing.
In order to enable GPU processing you should change the GPU attribute to true.
When GPU attribute is enabled the recipe installs a different version of TensorFlow plus CUDA and CuDNN libraries.

TensorFlow GPU support requires having a GPU card with NVidia Compute Capability (>= 3.0). More info here: CUDA GPUs

Environment

The following packages are automatically installed:

• Anaconda (python 3.5)
• Numpy
• Scipy
• Gym and its modules
• Universe and its modules
• Universe Starter Agent (AC3 Algorithm)
• Tensorflow 0.11
• Theano
• Keras
• OpenCV
• Docker
• Golang
• Cuda toolkit 8.0 + CuDNN 5.1 (if you choose to enable GPU processing)

Requirements

Platforms

• Ubuntu 14.04
• Ubuntu 16.04

Cookbooks

The following cookbooks are direct dependencies:

• apt '5.0'
• compat_resource

Chef

• Chef 12.1+

Recipes

This cookbook is split in 5 recipes in order to allow deployment on different setups and make it easy to split the job between several machines in distributed environments:

• Essentials:
  Installs basic packages

• Tensorflow:
  Install Anaconda/Python environment and Tensorflow 0.11

• Cuda(Optional for GPU processing):
  Install Cuda Toolkit and CuDNN

• Docker:
  Install Docker

• OpenAI:
  Install Gym, Universe and Universe Starter Agent

Authors

• Author:: Alexandro de Oliveira ([[email protected]](mailto: [email protected]))