Implementation of 'Exact indexing of dynamic time warping' - Keogh / Ann Ratanamahatana http://www.cs.ucr.edu/~eamonn/KAIS_2004_warping.pdf
A classifier of time series based on dynamic time warping is built. To improve performance Keogh Lower Bound and a classification system based on templates is implemented. As proved in the paper, these modifications make classifying a time series linear in complexity.
Main.trainDTW function receives the training set and the list of labels to detect. It returns a data structure composed of templates for each label. Given a new sample, it predicts its label by finding the most similar template.