Comprehensive Python module for computing and visualizing dynamic time warping alignment: DTWPy

Abstract

Dynamic Time Warping (DTW) is a well-known technique used to determine alignment between two temporal sequences. DTW has been used in wide range of applications and it can be applied on any data which can be represented as linear sequence. Existing DTW libraries have out dated implementation of core DTW algorithm, which result in low performance or are inapplicable for big sequences. The aim of this thesis is to present, a comprehensive DTW library, encapsulating t implementation of DTW variants with recently proposed efficient algorithms. In this thesis I presented a python module for computing and visualizing DTW alignment: DTWPy. DTWPy has implementation of classical DTW, almost all the DTW variants proposed in literature to date, with recently proposed performance efficient algorithms i.e. fastDTW and DDTW. Algorithms implemented in DTWPy are tested thoroughly and give correct results. Correctness is verified by comparing it with existing R implementation. Furthermore, I have compared algorithms implemented, and performance of DTWPy with existing libraries. DTWPy have the most comprehensive implementation of DTW algorithms present in literature to date, and is applicable on large temporal sequence. The architecture of DTWPy is designed to be flexible, to scale and accommodate possible future improvements.