Deep learning programming by all

We describe an open-source blocks-based programming library in Snap! [Harvey and Monig, 2010] that enables non-experts to construct machine learning applications. The library includes blocks for creating models, defining the training and validation datasets, training, and prediction. We present several sample applications: approximating mathematical functions from examples, attempting to predict the number of influenza infections given historical weather data, predicting ratings of generated images, naming random colours, question answering, and learning to win when playing Tic Tac Toe.