Tracing Tuples Across Dimensions A Comparison of Scatterplots

Tracing Tuples Across Dimensions A Comparison of Scatterplots and Parallel Coordinate Plots Xiaole Kuang (Master student, NUS) Haimo Zhang (Ph. D student, NUS) Shengdong (Shen) Zhao (Faculty member, NUS) (Faculty member, Michael J. Mc. Guffin École de technologie supérieure) 1

Welcome to The Last Talk of The Last Session of The Last Day! 2

of Vienna 9697 km Singapore 3

Systems, Tools, Interaction Techniques Vignette (CHI ‘ 12) Sand. Canvas (CHI ‘ 11) ear. Pod (CHI ‘ 07) Zone & Polygon Menu (CHI ‘ 06) MOGCLASS (CHI ‘ 11) Magic Cards (CHI ‘ 09) Elastic Hierarchy (Info. Vis ‘ 05) 4 Simple Marking Menu (UIST ‘ 04)

Visualization Techniques for Multi-Variate Data Parallel Coordinate Plot (PCP) Scatter Plot Matrix (SPLOM) Scatter Plot (SCP) 5

Why PCP vs. SCP? Both techniques are popular! Yet, we know very little about their comparative advantages. Viau et al. , TVGC 10 We need more systematic evaluations between PCP & SCP! Yuan et al. , TVGC 09 Claessen & van Wijk, TVGC 11 6

Basics of Evaluation Research question • What’s the comparative advantages between PCP & SCP for certain tasks? Task Independent variables Dependent variables 7

Basics of Evaluation Research question • What’s the comparative advantages between PCP & SCP for certain tasks? Task Independent variables Dependent variables 8

Basic Analytical Tasks Amar et al. : Low-level components of analytic activity in information visualization. Info. Vis 05, 111– 117. serves as a subtask for many other tasks (Holten & van Wijk, Euro. Vis 10) PCP is inferior than SCP (Li et al. , Info. Vis 10) 9

Value Retrieval Task Definition: • Given the numerical value of one attribute of a data tuple, find the numerical value of another attribute of the same data tuple. Multi-Variate Data Tuple (X 1, X 2, X 3, …. , Xn) a ? 10

Basics of Evaluation Research question • What’s the comparative advantages between PCP & SCP for certain tasks? Task Independent variables Dependent variables 11

Independent Variables Technique Parallel Coordinate Plot (PCP) X 1 X 2 X 3 X 4 SCP-rotated (Qu et al. , TVCG 07) X 2 X 2 X 1 X 3 X 4 4 3 X 2 X 3 X 1 SCP-common (SPLOM) X Scatter Plot (SCP) SCP-staircase 12 (Viau et al. , TVCG 10)

Independent Variable – 4 Technique PCP SCP-rotated (i. e. , Qu et al. , TVCG 07) SCP-common (i. e. , SPLOM) SCP-staircase (i. e. , Viau et al. , TVCG 10) 13

Additional Independent Variables Number of Dimensions X 1 X 2 X 2 X 3 X 3 X 4 X 5 X 4 Data Density X 1 X 2 X 3 X 4 14

Independent Variables • Technique • Dimension • Density 15

Dependent Variables • Completion time • Error distance 16

Experiment Demo 17

Experiment 1 Design 12 participants × 4 visualization techniques (PCP, SCP-common, SCP-rotate, SCP-standard) × 3 levels of data dimension (2 D, 4 D, 6 D) × 3 levels of data density (10 tuples, 20 tuples, 30 tuples) × 3 repetitions of trials = 1296 trials in total. 18

Overall Results Completion Time Error Distance Poorer Error Distance Seconds Poor Good Best PCP SCP-rotate SCP-common SCP-staircase Poor Good PCP SCP-rotate SCP-common SCP-staircase 19

st 1 Take-away Lesson PCP SCP-rotated (i. e. , Qu et al. , TVCG 07) SCP-common (i. e. , SPLOM) SCP-staircase (i. e. , Viau et al. , TVCG 10) 20

PCP vs. SCP-common 21

PCP vs. SCP-common Performance Difference Density 22

PCP vs. SCP-common Performance Switch Order Density 23

Important Observation There seems to be a Density & Number of Dimension Trade-off between PCP & SCP-common! 24

Experiment 2 × 18 participants × 2 techniques (PCP, SCP-common) × 3 dimensions (4 D, 6 D, 8 D) [2 D, 4 D, 6 D in Exp. 1] × 3 densities (20 tuples, 30 tuples, 40 tuples) [10, 20, 30 in Exp. 1] × 5 trials for each combination = 1620 trials in total. 25

Results – Completion Time Overall result for Exp. 2 faster SCP-common (15. 41 s) PCP (18. 23 s) Result in Exp. 1 SCP-common (12. 02 s) faster PCP (8. 99 s) Trade-off between number of dimensions & data density Dimension Density 26

Results – Error Distance Trade-off between number of dimensions & data density Dimension Density 27

Take-away Lessons Dimension The value retrieval performance of PCP increases depending on dimensionality. The performance of SCPcommon seems independent of dimensionality. Increasing density affects the performance of PCP more than it affects SCP-common. Density 28

Let’s Recap the Take Away. Messages and Ask Why 1) Both SCP-rotate and SCP-staircase are inferior for value retrieval task 29

Let’s Recap the Take Away Messages 2) Performance trade-off between PCP & SCP-common for both dimensionalities and data density. • PCP increases depending on dimensionality. • SCP-common performance seems to be independent. 30

Let’s Recap the Take Away Messages 2) Performance trade-off between PCP & SCP-common for both dimensionalities and data density. • PCP increases depending on dimensionality. • SCP-common performance seems to be independent. • Increasing density affects the performance of PCP more than it affects SCPcommon. 10 tuples 40 tuples 31

Conclusion and Future Work Our study helps to understand the comparative advantages between PCP & SCP However, this is only a starting point, 32

The Grand Vision Ideally, this problem can be solved by … Info. Vis evaluation package Results/ Recommendations 33

Acknowledgment This research is supported by: The National University of Singapore Academic Research Fund R-252 -000 -375 -133 and by: The Singapore National Research Foundation under its International Research Centre @ Singapore Funding Initiative and administered by the IDM Programme Office.

Q&A Elastic Hierarchy (Info. Vis ‘ 05) Tracing Tuples Across Dimensions (Euro. Vis ‘ 12) 35

End 36