Parallel Coordinates

1
Parallel Coordinates

• Representation of multi-dimensional data
• Discovery Process
• xmdv Visualization Tool

Ganesh K. Panchanathan Christa M. Chewar
2
Cartesian Vs. Parallel Coordinates
1500
1500
1500
Price()
0
0
Speed( Mhz) 1500
Speed(Mhz) Price()
3
473 Items, 16 Dimensions
X1 Yield X2 Quality X3 - X12 Defects
X13 - X16 Physical Parameters
4
The Discovery Process

• Identify and understand Objectives
• Combine atomic queries to form complex queries

Isolate batches with high X1 and X2
Batches with low X3 do not have high yield and
quality
5
Isolate batches with Zero Defects in 9 attributes
All 9 batches have poor yield, quality Process
sensitive to changes in X6
Isolate batches with Zero Defects in 8 attributes
Small amounts of X3 and X6 defects necessary for
high yield and quality
6
Further Insights
Higher Range of split in X15

• Low Yield
• Inconsistent Quality

Lower Range of split in X15

• High Yield
• High Quality

7
Conclusion
To get high yield and quality
Small Ranges of defects X3 and X6 are necessary
Lower range of physical parameter X15
8
XMDV Tool
9
Grading Data
10
Pros

• Simplicity in Representation ( x D ? 2 D)
• Scalability ( any N)
• Visual cues from items having similar properties
• Uniform treatment of all variables
• Finds relationships between variables
• Combine atomic queries to form complex query

Cons

• Difficulty with large data sets