This is a short tutorial involving writing and training an autoencoder from scratch on the MNIST dataset using autograd. Part of the git book for the PIGlets reading group at the University of Edinburgh.

The requirements file is defined as a conda yml file. The easiest way to install the environment is to use miniconda, and the fastest way to install miniconda is on a 64 bit Linux system is:

curl https://repo.continuum.io/miniconda/Miniconda-latest-Linux-x86_64.sh | bash

Once installed, you can then install an environment with this yml file:

conda env create -f environment.yml

Then, to activate it:

source activate autoencoders

Everything is in the Jupyter notebooks, so start up the Jupyter server with:

jupyter notebook

The server will automatically open in your browser (or you'll find it at http://localhost:8888) and you can open the notebook files.