Why do you care? Time-series data is all over the place. - PDF document

Why do you care? • Time-series data is all over the place. Time-Series Data Kaitlin Duck Sherwood CS 533c What is Time-Series Data? What is Time-Series Data? • Lines. • Usually periodic Source: van Wijk and van Selow, Cluster and Calendar based Visualization of Time Series Data, 1999 Spiral Viewer (Carlis et al) Unknown periodicity • Angle � position in cycle Color, diameter • Tweak period in realtime to find available for use • Radius � cycle number periodicity • Example: music 1

Helices Spiral barchart • Example: Chimp eating Z-axis used for data habits • Angle � day of year • Radius � year • Z-axis* � type of food • Color � type of food • Diameter � amount Can step through spokes one at a time (Rings to beat occlusion) User feedback Good points of paper • Qualitative feedback from 12 • Compelling visuals users • Gave examples • “Buy into the notion of a spiral • Has software display” • Some user feedback • Couldn’t self-operate • Wanted more Bad points of paper Time-series bitmaps • Examples not compelling Repurposes heavily: • Graphs unlabeled • Chaos Game • Difficult to see quant info Representation • Questionable movie data • SAX • Weak user eval • Windows Explorer • Advantages over Cartesian? 2

Chaos Game Representation SAX • Assign corner of square to each base • Converts reals into equiprobable letters • For each symbol, take a step in symbol • Eliminate trends with narrow window direction of half distance • Uses: clusters, motifs, anomalies • Color corresponds to number of times a pixel visited Time-series bitmaps Ubiquity • Data -> SAX -> CGR -> bitmap Use filesystem: • Linear color mapping (JET) • Thumbnails • Length normalization • Cluster (using MDS) Comparisons only Real-world data Good points • Clustering heterogenous data (15 • Pretty cool idea sets) • Repurposes material from other – Better than ARIMA or Markov fields well • Clustering of 20 ECG patients (perfect) • Ubiquitous visualization (filesys) • Video classification • Impressive results – Better than Euclidian or DTW • Classifying ECG data (perfect) 3

Bad points Summary • Didn’t explain CGR well • Spirals. Cool pictures, what use? • Didn’t explain Windows • Bitmaps. Less cool, perhaps clustering more useful. • Data sets relatively small • No user testing References • Interactive Visualization of Serial Periodic Data , John V. Carlis and Joseph A. Konstan, Proc UIST 98. • Time-series Bitmaps: A Practical Visualization Tool for working with Large Time Series Databases Kumar, N., Lolla N., Keogh, E., Lonardi, S. , Ratanamahatana, C. A. and Wei, L. (2005). Proc. SDM '05, pp. 531-535 4