As part of my PhD project I am estimating personal energy expenditure from smartphone data were one of the modalities are accelerometry. The literature suggests that estimation of energy expenditure is improved if conditioned on physical activity type why a first step was to classify human activity from accelerometer data.
This repo is a compilation of data processing scripts, model definitions and support code to train the models. Dependencies include Theano and Lasagne.