This project began as part of the Studio 415 ML Meetup. I am forever grateful Steven Webster and his XD team for providing the space we used when we were first meeting.

The bash script grabDependencies.sh will download data and clone several repositories that contain examples and dependencies. It will create a directory data and subdirectories to contain the MNIST, CIFAR-10 and CIFAR-100 datasets.

The python scripts in the code directory expect these directories to exist, as well as an additional directory, data/cifarKaggle, which must be populated by hand using the data from Kaggle.com. Expected files and subdirectories are

• data/cifarKaggle/trainLabels.csv from http://www.kaggle.com/c/cifar-10/download/trainLabels.csv
• data/cifarKaggle/train the unzipped contents of http://www.kaggle.com/c/cifar-10/download/train.7z
• data/cifarKaggle/test the unzipped contents of http://www.kaggle.com/c/cifar-10/download/test.7z

The python scripts in the code directory will try to identify the location of the cifar-ten directory. First they will check if there is an environment variable CIFAR10_HOME, in which case they will assume that is the location of the cifar-ten directory. If that environment variable is not set, then it will look for python scripts in directory $HOME/cifar-ten/code/ and data files in subdirectories of $HOME/cifar-ten/data.

Currently the python scripts in code require packages numpy, scipy, Theano and PyWavelets. To install these packages system-wide, invoke

• sudo pip install Theano
• sudo pip install pywavelets --no-deps

If local installation is sufficient or preferred, invoke pip with the --user flag.