mila-iqia / summerschool2015

Slides and exercises for the Deep Learning Summer School 2015 programming tutorials

summerschool2015

This repository contains the slides and exercises for the Deep Learning Summer School 2015 programming tutorials.

Installation instructions

The tutorials are written in Python, using Theano and Fuel. They are designed to be run locally on a laptop, without using a GPU.

Python and dependencies

The simplest way to install a Python software stack with most dependencies is to use Anaconda.

First, download and execute the installer. You can install it as a user (you do not have to use sudo). We recommend that you let the installer make Anaconda the default Python version.

Then, in a terminal:

$ conda update conda

Additional steps for Windows

These additional steps are required for Windows:

• Download Git, and execute the installer. This will be necessary to get the latest version of Theano and Fuel. We recommand you select "Use Git from the Windows Command Prompt" option, so you can execute all the following command lines from the regular Windows cmd shell.

• Install a C++ compiler and Python DLL. From a shell:

  conda install mingw libpython

Optional: Additional step to display the graphics

If you do not follow these steps, the pydotprint command will raise an exception and fail, but the other functionalities of Theano would still work.

On Ubuntu/Debian

$ sudo apt-get install graphviz
$ conda install pydot

On Fedora, CentOS, Red Hat Enterprise

$ sudo yum install graphviz
$ conda install pydot

On MacOS

• Download graphviz from http://www.graphviz.org/Download_macos.php

• Then, from a terminal:

  $ conda install pydot

On Windows

• Download graphviz from http://www.graphviz.org/Download_windows.php

• Add to the PATH environment variable the directory where the binaries were installed, by default C:\Program Files (x86)\Graphviz2.38\bin

• Then, from a terminal:

  pip install pydot_ng

Optional: MKL Blas

If you are eligible to an academic license for Anaconda add-ons, you can download and install the MKL optimizations. This will bring a small speed improvement for dot products, but is not critical for the tutorials at all. Once you have obtained the license:

$ conda install mkl

Theano

There have been some improvement and bug fixes since the last release, so we will use the latest development version from GitHub. The following command installs it for the current user only:

$ pip install git+git://github.com/Theano/Theano.git --user

Note

If you are using Windows and selected "Use Git from Git Bash only" when installing Git, or if the command above failed because git is not available in the path, then you need to run the command line above from the "Git Bash" terminal instead of the regular Windows command prompt.

If you are following these instructions in advance, you may need to execute this command in order to get last-minute fixes:

$ pip install --upgrade --no-deps git+git://github.com/Theano/Theano.git --user

Note

If you install Theano for the current user only (using --user), command-line utilities (for instance theano-cache) will not be accessible from a terminal directly. You would have to add the script installation directory to the PATH environment variable.

• On Mac OS and Linux, that path is $HOME/.local/bin by default.
• On Windows 7, that path is C:\<User>\AppData\Roaming\Python\Scripts if your user name is "<User>".

Fuel

We install the development version of Fuel from GitHub.

$ pip install git+git://github.com/mila-udem/fuel.git --user

If you are following these instructions in advance, you may need to execute this command in order to get last-minute fixes:

$ pip install --upgrade --no-deps git+git://github.com/mila-udem/fuel.git --user

Note

If you install Fuel for the current user only (using --user), command-line utilities (for instance fuel-download and fuel-convert) will not be accessible from a terminal directly. Unless you have already performed that step when installing Theano, you would have to add the script installation directory to the PATH environment variable.

• On Mac OS and Linux, that path is $HOME/.local/bin by default.
• On Windows 7, that path is C:\<User>\AppData\Roaming\Python\Scripts if your user name is "<User>".

Get and run these tutorials

First, clone this repository:

$ git clone https://github.com/mila-udem/summerschool2015.git

To use the IPython notebooks, you have to launch the IPython server on the base directory:

$ ipython notebook summerschool2015

A new window or tab should open in your web browser. If it does not (or if you want to use it in a different browser), the previous command should mention a URL you can open, probably http://localhost:8888/. From there, you can navigate to the .ipynb files.