Interactive Pattern Search in Time Series (Using TimeSearcher 2)

1
Interactive Pattern Search in Time Series(Using
TimeSearcher 2)

• Paolo Buono, Aleks Aris, Catherine Plaisant,
• Amir Khella, and Ben Shneiderman
• Proceedings, Conference on Visual Display and
  Analysis, 2005

Presented to CMSC 838S (Information
Visualization) By Derek Juba Scott Nestler On
Feb. 16, 2006
2
Outline

• Introduction
• Related Work
• TimeSearcher 2 Demo
• Key Contributions
• Future Work
• Conclusions

3
Introduction

• Time series- sequence of real numbers,
  representing observations of variable at equal
  time intervals
• Many applications of time series (e.g. EKGs,
  seismographs, digital recordings)
• Traditional analyses are based in classical
  statistics
• Now able to explore time series data with
  visualization tools
• TimeSearcher 2 designed for range of users
  statistical analysis skills not required

4
Introduction (cont)

• Timebox- rectangular region that is selected and
  directly manipulated on a timeline overview of
  data
• Boundaries of timeboxes used to specify
  parameters for query
• Two types
• Original- used to filter data and reduce scope of
  search
• New- used to perform specific pattern search in
  remaining data

5
Related Work

• Diamond Fast (Unwin Wills, 1999)
• Allows visualization, moving, resizing
• Only manages short time series
• ILOG Personal Stock Monitor
• High level of interaction with enhanced zoom
• Lack search capabilities limited to a single
  time series
• Semantic zoom (Brodbeck, Girardin, 2003)
• Spiral view(Carlis Konstan, 1998)

6
Related Work (cont)

• Time Searcher 1- patterns specified with
  timeboxes
• Choratas- pattern specified numerically
• VizTree- pattern specified in segments
• QuerySketch- allows direct sketching of pattern
• IPBC- 3D tool allows selecting pattern in data
  for periodic data

No known tool other than TimeSearcher allows
specifying multiple patterns on multiple variables
7
TimeSearcher 2 Demo
Demo
8
Key Contributions

• Three-step interactive search
• Reduce scope with timebox(es)
• Specify a pattern to search for
• Refine query dynamically
• Search algorithm
• Uses modified Euclidean distance between two
  time series
• Transformations available
• Offset translation
• Magnitude scaling
• Linear trend removal
• Noise reduction

9
Future Work

• Dealing with larger datasets
• Improved interaction
• Search pattern selection
• Dealing with missing data
• Ignore?
• Estimate?
• Evaluation

10
Conclusions

• Room for improvement in interactive exploration
  of time series
• Improved algorithms
• Interactive interfaces
• Interactive search used with traditional features
  aids exploratory analysis
• Three-step framework applicable to larger data
  archives