VizDB - People
Download
Report
Transcript VizDB - People
VizDB
A tool to support
Exploration of large databases
By using
Human Visual System
To analyze mid-size to large data
Data Mining Techniques
Implements several data mining techniques
Pixel-oriented Techniques
(Spiral, Axes, and
Grouping Techniques)
Parallel Coordinates
Stick Figures
Exploration of unto a million data values
Concept
The basic idea for visualizing the data is
to map the distances to colors and
represent each data value by one or
multiple colored pixels.
Interactivity is the key !
Requirement
Feedback required when query returns
unexpected results
Interactivity allows immediate feedback
from a modified query
Configurable tool, that allows various forms
of data visualization techniques
Using the human vision system for pattern
recognition
Basic Technique
Sort query data w.r.t. the relevance and map
relevance factors to colors
Highest relevance factor in the center
Yellow-Green-Blue-Red-Black in decreasing order
of relevance.
Separate window for each selection predicate in
the query
Multiple windows make multi-dimensional
visualization
Mapping 2-D To The Axes
Visualization of inherently 2D or 3D data is
not dealt with in VizDB
Where no inherent 2D semantics of data
exist, VizDB is a valuable tool. Use of two
axes for two dimensions. Positive as well as
negative values displayed.
Some space may be wasted .. (Why?)
Grouping
Each area is arranged in a rectangular spiral
shape according to relevance factors
Coloring is similar to the previous method
Grouping allows data similar in one
dimension to be grouped together. Data in
multiple dimensions are represented as
clusters of pixels
Good for larger dimensionality
Interactive Data Exploration
Dynamic Query Modification Techniques
Feedback on the results
– Change in color means change in values that are
“relevant”
– Change in structure means overall distribution of data
has changed
Sliders for discrete as well as continuous values
Initial Query is SQL or “Gradi”
Calibrations
Calculation of “relevance” factor can be
calibrated by the user
Starting and ending values for various
numeric data
– Eg: Blood samples count
What about complex queries?
Multiple layers of windows for complex
queries using nested AND and OR operators
Data that satisfies ALL joins is yellow. The
rest is colored according to number of
criteria met
Works well with the relational databases
Implementations
C++ with Motif using X Windows on HP
7xx
Currently being ported to Linux (I couldn’t
get this working! )
Adding new techniques
More Info Viz. Techniques can be integrated
with the system.
New
Latest version supports Parallel coordinates, Stick Figures, Pan and zoom Stuff !!
techniques
Applications
Molecular Biology - to find possible
docking regions by identifying sets
surface points with distinct
characteristics.
Database of geographical data
Environmental Data
NASA Earth observation data
Future Work
Automatic generation of queries that
Cool !!
correspond to data in specific regions
(Select some data, and the SQL query that
matches that data will get generated..
Time series visualization
Thank You
The presentation slides are available at
http://filebox.vt.edu/users/adatey/research/Viz
DB.ppt
A small color picture that shows different
techniques
http://filebox.vt.edu/users/adatey/research/Vis
DBHandout.eps