Magic Beans for Deep Learning

Setting up your developer environment can be a pain, this post give you a handful of bean grow beanstalks into your very own deep learning castle in the (Amazon) cloud. Now, there are a number of tutorials that walk you through the steps of getting one or more libraries set up, here are two great ones:

However, most deep learning installer tutorials are lacking in several respects and the rest don't really work (or skip steps). Thus the need for this anti-tutorial that simply gives you the magic one-liner (a-la installing brew) to get all the common libraries setup and configured (on the lastest P2 class machines running Ubuntu 16.04) in Amazon's EC2..

The incantation below will give you everything you could possibly need, including

  1. Latest CUDA 8
  2. Latest CuDNN 5.1
  3. MXNet
  4. Torch
  5. Theano
  6. TensorFlow
  7. Keras
  8. OpenCV & DLib

For the impatient I've made a public image (ami-be391da9) on AWS. Alright, (drum roll please) here's the single line install command you can simply paste it into the terminal:

wget http://public-fileshare.s3.amazonaws.com/install.gz; tar xzf install.gz; rm install.gz; cd scripts; chmod +x master_installer.sh;./master_installer.sh

Should take about 2 hours to complete. But once it's done, you can check that everything works:

As you can see, torch and all the python imports work, but please let me know if the comments if anything breaks for you. Happy coding!