Deep learning examples

This repository is a supplement to my class Machine Learning and Big Data taught at the CUNY MS Data Analytics program. More examples will be added over time.

Installation

The examples expect a Linux environment with Docker installed. If you meet this requirement, you can simply use the Makefile to build a local Docker image and run a container. These instructions assume that you've downloaded the repository to ~/workspace/deep_learning_ex. If you clone to a different directory, you will need to change the Makefile.

First build the container with sudo make. This will download a Docker image and then install some additional packages. This image will be saved locally as zatonovo/dlx.

Now you are ready to run a container based on the image. You have the option of starting a bash shell, a Torch session (in Lua), or a Python session.

sudo make bash
sudo make torch
sudo make python

When you start the session, you'll see that all the files in the host directory ~/workspace/deep_learning_ex are visible in the current working directory of the image! This is the beauty of volume mapping. This means that any data and code you create in the container will be saved in your host file system to use for further analysis. For example, if you want to visualize your results in R, it's easier to use your existing R configuration and load the data files, as opposed to running R from inside the container.

Examples

Function approximation

This example approximates a bivariate continuous function in Torch. For background, refer to this primer.

To run the example, start a Torch session, as instructed above. You should see the standard Torch prompt inside Lua.

  ______             __   |  Torch7                                         
 /_  __/__  ________/ /   |  Scientific computing for Lua. 
  / / / _ \/ __/ __/ _ \  |  Type ? for help                                
 /_/  \___/_/  \__/_//_/  |  https://github.com/torch         
                          |  http://torch.ch                  
	
th> 

From this prompt, type dofile 'ex_fun_approx.lua', which will run the script. The script itself creates only the most basic of networks. To get results similar to the end of the primer, you need to add appropriate layers and activation functions to the network. You also need to specify an appropriate cost function and tune the parameters of the optimizer. These are all exercises for my students/readers.

Resources

Torch

The Torch project has excellent documentation. This example provides a full walkthrough of setting up a neural network and training it.