Incvisage：逐步改进可视化以支持快速决策

1 .I’ve Seen “Enough” Incrementally Improving Visualizations to Support Rapid Decision Making Sajjadur Rahman , Maryam Aliakbarpour , Ha Kyung Kong , Eric Blais , Karrie Karahalios , Aditya Parameswaran , Ronitt Rubinfield University of Illinois (UIUC) , MIT , University of Waterloo

2 .Data Visualization 2 Microsoft’s Power BI hits 5m subscribers Tableau software soars 17% on Q3 earnings

3 .Visualiz ing Large Scale Data Key challeng e : reveal insights within seconds! 3 Even delays > 0.5 seconds hamper insights! [Liu et al. 2015] Visualization ~60s Time Delay US Flight Data (74m data points)

4 .Motivation Faster analysis of large datasets Time series/Ordinal datasets  (trendlines, heatmaps) 4 Trendlines Heatmaps

5 .Motivation Faster analysis of large datasets Time series/Ordinal datasets  (trendlines, heatmaps) 5 Time Avg. Delay How can we generate visualizations faster?

6 .How can we generate visualizations faster? Pre-computation data cubes ( Nanocube , Profiler, Immense) Dimensionality Reduction (M4) Sampling 6

7 .Sampling: Existing Solutions Incremental OLA (online aggregation) sampleAction 7

8 .Sampling: Existing Solutions Incremental OLA (online aggregation) sampleAction 8

9 .OLA Approach (2) Standard approach: sampling based incremental visualization 9 t 1 t 2 … t 4 … t 7 … Completion

10 .OLA Approach ( 2 ) 10 t 1 t 2 … t 4 … t 7 … Completion Standard approach : sampling based incremental visualization

11 .OLA Approach ( 2 ) 11 t 1 t 2 … t 4 … t 7 … Completion Standard approach : sampling based incremental visualization

12 .OLA Approach ( 2 ) 12 t 1 t 2 … t 4 … t 7 … Completion Standard approach : sampling based incremental visualization

13 .OLA Approach ( 2 ) 13 t 1 t 2 … t 4 … t 7 … Completion Standard approach : sampling based incremental visualization

14 .Motivation (2)  Hard to draw conclusions 14  Potentially incorrect  Jarring experience/unstable t 1 t 2 … t 4 … t 7 … Completion Standard approach: sampling based incremental visualization Can we do any better?

15 .Incrementally Improving Visualizations 15 Standard Incrementally Improving t 1 t 2 … t 4 … t 7 … Completion Show an increment when confident

16 .Incrementally Improving Visualizations 16 Segment Keep on sampling until confident t 1 t 2 … t 4 … t 7 … Completion

17 .Incrementally Improving Visualizations 17 2-Segment Approximation, L 2 t 1 t 2 … t 4 … t 7 … Completion Keep on sampling until confident

18 .Incrementally Improving Visualizations 18 t 1 t 2 … t 4 … t 7 … Completion k -segment approximation at iteration k

19 .I ncrementally Improving Visualizations 19 emergent features t 1 t 2 … t 4 … t 7 … Completion

20 .Incrementally Improving Visualizations 20 t 1 t 2 … t 4 … t 7 … Completion

21 .Incrementally Improving Visualizations 21 t 1 t 2 … t 4 … t 7 … Completion Draw conclusions with guarantees

22 .22 Incrementally Improving Visualizations Rapid decision making! Incrementally Improving Standard + insights with guarantees - no explicit guarantees on insights + important features first - no notion of importance + incremental improvement + incremental improvement

23 .23 Incrementally Improving Visualizations Any negatives? Incrementally Improving Standard + insights with guarantees - no explicit guarantees on insights + important features first - no notion of importance + incremental improvement + incremental improvement

24 .24 Incrementally Improving Visualizations Any negatives? Incrementally Improving Standard + insights with guarantees - no explicit guarantees on insights + important features first - no notion of importance + incremental improvement + incremental improvement

25 .25 Incrementally Improving Visualizations Incrementally Improving Standard + insights with guarantees - no explicit guarantees on insights + important features first - no notion of importance + incremental improvement + incremental improvement - information hidden + display all the information

26 .26 Incrementally Improving Visualizations Incrementally Improving Standard + insights with guarantees - no explicit guarantees on insights + important features first - no notion of importance + incremental improvement + incremental improvement - information hidden + display all the information

27 .Challenges Algorithmic how does the algorithm work? Interface how do the users interpret the visualizations? 27

28 .Outline Incrementally Improving Visualizations Problem Statement ISplit Sampling Technique System Architecture Performance Evaluation and User Study Discussion 28

29 .Problem Statement: Query Type Time ( X a ) Delay ( AVG (Y) ) X a AVG(Y) X b