DTW Motif Discovery

This website includes all the codes, data, experiments and the figures generated for the paper "Exact Discovery of Time Series Motifs under DTW".

What did we do? We introduced a practical tool, which we call SWAMP (Scalable Warping Aware Matrix Profile ), to find DTW-based motifs in large datasets. Our method automatically performs the best trade-off between time-to-compute and tightness-of-lower-bounds for a novel hierarchy of lower bounds representation we introduce. We show that under realistic settings, our algorithm can admissibly prune up to 99.99% of the DTW computations.

Why use DTW? In many domains, DTW-based motifs represent semantically correct conserved behavior that would escape our attention using all existing Euclidean distance-based methods. Here are some examples:

The expanded version of the paper which includes all the details, experiments and case studies can be found here.

Datasets

All datasets used for this paper (with the reference to the figures in the paper) are available in the following links :

Dishwasher (Figures 1-3)
EPG data collected from an Asian citrus psyllid (ACP) (Figure 12)
Seven-hour segment EPG data collected from an ACP (Figure 13)
The electrical power demand of a UK house over 18.5 days (Figure 14)
Left-hand movement (Figure 15)
Hip-worn Gyroscope (Figure 16)
Cricket umpire signs (Figure 17)
Telemetry from a wild penguin hunting at sea (Figure 18)
Melbourne Pedestrian Traffic (Figure 20.left)
Respiration Signals (Figure 20.right)
The Reproducibility “Rosetta Stone” (Figure 21)

Codes

All codes in this project have been implemented in Matlab, and are accessible via the following link:

DTW Motif Discovery Codes

Figures

Large scale versions of the figures in the paper are available in the following PowerPoint slides:

DTWMotif.pptx

Supplementary materials

Additional tests are provided in the following subpages:

Scalability test results are available here.

Page updated

Google Sites

Report abuse