README.md

GPU development environment setting

This is for Ubuntu16.04 with cuda8.0. Using python library, Theano, Lasagne, and libgpuarray to implement deep learning with Nvidia GPU. Note that, all tool's version are important!

Dependencies

• Ubuntu 16.04
• NVIDIA GPU
  • Cuda 8.0
  • cuDNN 5.0
• Deep Learning tool
  • Theano 0.9
  • libgpuarray, pygpu 0.7.3
  • Lasagne 0.1
  • Arcade-Learning-Environment
  • Pylearn2
• Deep Learning Project
  • deep_q_rl
  • Q-Optimality-Tightening

Because Theano now only supports to cuDNN with version 5.1. And the newest Theano version needs pygpu with version 0.7.0 or newer, for now, can not install gpuarray from github. Please make sure the version of Theano is compatible with the one of Lasagne!

Install

First, install cuda and cuDNN. Secondly, install Theano, Lasagne and gpuarray.

Run

Finally, testing with Deep Learning projects.

• Lasagne MNIST
• Q-Optimality-Tightening
• deep_q_rl