Professional Documents
Culture Documents
Importance-Driven Time-Varying Data Visualization: Chaoli Wang, Hongfeng Yu, Kwan-Liu Ma University of California, Davis
Importance-Driven Time-Varying Data Visualization: Chaoli Wang, Hongfeng Yu, Kwan-Liu Ma University of California, Davis
Data Visualization
Chaoli Wang, Hongfeng Yu, Kwan-Liu Ma
University of California, Davis
Importance-Driven Volume Rendering
• Block-wise approach
• Importance evaluation
• Amount of information a block contains by itself
• New information w.r.t. other blocks in the time series
• Information theory
• Entropy
• Mutual information
• Conditional entropy
Information Theory
• Entropy
H ( X ) p( x ) log p( x )
xX
• Mutual information
p( x, y )
I ( X ; Y ) p( x, y ) log
xX yY p( x ) p( y )
• Conditional entropy
H ( X | Y ) H ( X ) I ( X ;Y )
H(X) H(Y)
• Feature vector f3
• Data value f2
• Gradient magnitude or other derivatives
• Domain-specific quantities
f1
• Multidimensional histogram
• Use the normalized bin count as probability p(x)
H ( X ) p ( x ) log p( x )
xX
Importance in Joint Feature-Temporal Space
F
• Use joint feature-temporal histogram
• Use the normalized bin count as probability p(x,y)
• Run-length encode the histogram
p( x, y )
I ( X ;Y ) p ( x, y ) log
xX yY p( x ) p( y )
Importance Value Calculation
I regular
T
Importance Curve – Climate Data Set
I periodic
T
Importance Curve – Vortex Data Set
I turbulent
T
Clustering Importance Curves
90 time steps
90 segments
Cluster Highlighting – Earthquake Data Set
Cluster Highlighting – Hurricane Data Set
Cluster Highlighting – Climate Data Set
Cluster Highlighting – Vortex Data Set
Cluster Highlighting – Combustion Data Set
Abnormality Detection
A: El Niño B: La Niña
Time Budget Allocation
• Animation time
• Non-even allocation
• Rendering time
• Assign to each time step (and each block in a time step)
• Adjust the sampling spacing accordingly
Time Step Selection
• Uniform selection
• Importance-driven selection
• Select the first time step
• Partition the rest of time steps into (K-1) segments
• In each time segment, select one time step: t arg max H ( | t )
• Maximize the joint entropy
Precomputation and Clustering Performance
• The test data sets with their parameter settings, sizes of joint feature-temporal histograms,
and timings for histogram calculation.
• Timing for clustering all time steps of the five test data sets.
Choices of Window and Bin Sizes
• The importance curve of the vortex data set with different time window sizes (W) and
numbers of bins for feature components F = (f1, f2, f3).
Choices of # of Clusters and Block Size
• The cluster of the highest importance values under different choices of number of clusters
and block size. Top row: color adjustment only. Bottom row: color and opacity adjustment.
Artifact Along Block Boundaries
20×20×20 10×10×20
Summary
• NSF
• CCF-0634913, CNS- 0551727, OCI-0325934, OCI-0749227, and
OCI-0749217
• DOE SciDAC Program
• DE-FC02-06ER25777, DE-FG02-08ER54956, and DE-FG02-
05ER54817
• Data sets
• Combustion: Jacqueline H. Chen, SNL
• Climate: Andrew T. Wittenberg, NOAA
• Earthquake: CMU quake group
• Hurricane: NSF, IEEE Visualization 2004 Contest