This repository contains the slides and exercises for the Theano tutorial at the Deep Learning School in Stanford, September 24-25, 2016.

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

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

For Windows users, please use the Python 2.7 version (or use a conda environment with Python 3.4), as Theano does not support Pyton 3.5 on Windows yet.

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

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

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.

$ sudo apt-get install graphviz
$ pip install pydot-ng

$ sudo yum install graphviz
$ pip install pydot-ng

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

• Then, from a terminal:

  $ pip install pydot-ng

• 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

There have been some improvement and bug fixes since the last release, so we will use the latest development version from GitHub.

Use --user if you installed Anaconda for all users and only want to install Theano for the current user.

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

The LSTM tutorial relies on Fuel for on-the-fly data processing. We install the development version of Fuel from GitHub.

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

If the command above failed with

.. ERROR:: Could not find a local HDF5 installation.
You may need to explicitly state where your local HDF5 headers and
library can be found by setting the ``HDF5_DIR`` environment
variable or by using the ``--hdf5`` command-line option.

run the following command and try again

$ conda install h5py pytables

Or, on MacOS

$ brew install homebrew/science/hdf5

First, clone this repository:

$ git clone https://github.com/lamblin/bayareadlschool.git

To use the Jupyter Notebooks, you have to launch the Jupyter server on the base directory:

$ jupyter notebook bayareadlschool

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.