OutlierPreserving FocusContext Visualization in Parallel Coordinates Matej Novotn

Outlier-Preserving Focus+Context Visualization in Parallel Coordinates Matej Novotný Helwig Hauser Comenius University Bratislava, Slovakia VRVis Research Center Vienna, Austria

Our goal § A parallel coordinates visualization that: § Employs Focus+Context § Handles outliers § Renders effectively 2 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Overview § Motivation § Workflow § § § Benefits Bonus! Results and conclusions 3 § Abstraction, Focus+Context § Outliers § Binning § Context Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Parallel Coordinates § § 4 Insight into multidimensional data Correlations, Groups, Outliers Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Parallel Coordinates § § 5 Insight into multidimensional data Correlations, Groups, Outliers Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Parallel Coordinates § § 6 Insight into multidimensional data Correlations, Groups, Outliers Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Large data visualization § Large data cause clutter in visualization § 16. 000 records 7 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Large data visualization § Transparency used to decrease clutter § 16. 000 records 8 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Large data visualization § Transparency used to decrease clutter ? § 32. 000 records 9 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Large data visualization § Transparency used to decrease clutter ? ? § 64. 000 records 10 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Large data visualization § Transparency used to decrease clutter ? ? ? § 100. 000 records 11 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Large data visualization § Transparency used to decrease clutter ? ? ? § Do these records belong to the main trend? 12 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Data abstraction § § 13 Density-based representation of data 16 bins Trends are clearly visible Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Related work § Hierarchical Parallel Coordinates (Fua et al. , 1999) § Visual representation of clusters § Smooth transparency § Cluster centers emphasized 14 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Related work § Revealing Structure within Clustered Parallel Coordinates Displays (Johansson et al. , 2005) § Textures, density § Transfer functions § Clusters § Outliers 15 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Outliers § Different, sparse, rare § Why should we care? § Outliers can be considered in: 16 § Investigation (special cases in simulations…) § Security (network intrusion, suspicious activity…) § Detect errors in data acquisition § Data space § Screen space Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Outliers are like kids. If you leave them unattended they either get lost or they break stuff. 17 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Outliers § Avoid losing them in visualization § Improve data abstraction or F+C 18 § e. g. due to transparency or abstraction § e. g. remove outliers from clustering Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Workflow 19 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Workflow 20 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Step 1: Binning § 2 D binning § Density-based rep. § Screen-oriented § Low memory demands compared to n. D § Every segment separately § Result = bin map 21 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Benefits of binning? § § § Operations no longer depend on the size of the input Information is preserved Variable precision of binning § Variable precision of § § 22 visual output Fine binning does not destroy details Larger bins can be produced from finer bins Matej Novotný http: //www. VRVis. at/ 128 x 128 bins Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Step 2: Outlier detection § Various criteria can be employed § e. g. isolated bins, median filter … 64 x 64 bin map 23 Matej Novotný http: //www. VRVis. at/ 32 x 32 bin map median filter 32 x 32 bin map isolated bins Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Step 3: Generating Context § § Outliers → opaque lines Binned trends → quads § Population mapped to color intensity § No blending § Low visual complexity § Rendering order according to population 24 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates 8 bins

Step 4: Add Focus 8 bins 25 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Benefits § Operations performed on bin maps § Outliers handled separately § Output-sensitive implementation 26 § Reduced complexity § Results coherent with visual output § More operations feasible – e. g. Clustering § Increased information value § Clearer context § View divided into layers and segments Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Results § Large data can be rendered and explored § 3 millions records, 16 dimensions, 32 bins 27 § Binned in 30 sec, rendered instantly (3 Ghz, 64 bit) Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

BONUS! Clustering 28 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Clustering, step 0 § Apply Gaussian to smooth out the bin map § Segmentation data, Green vs Darkness 29 Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Clustering, further steps § § Start with the highest population Decrease the population threshold § Old clusters grow § New clusters emerge 50% 30 Matej Novotný http: //www. VRVis. at/ 20% 10% Outlier-Preserving Focus+Context Visualization in Parallel Coordinates 0%

Clustering results R 31 Matej Novotný http: //www. VRVis. at/ B G Outlier-Preserving Focus+Context Visualization in Parallel Coordinates D S H

Clustering results R 32 Matej Novotný http: //www. VRVis. at/ B G Outlier-Preserving Focus+Context Visualization in Parallel Coordinates D S H

Clustering results R 33 Matej Novotný http: //www. VRVis. at/ B G Outlier-Preserving Focus+Context Visualization in Parallel Coordinates D S H

Clustering results R 34 Matej Novotný http: //www. VRVis. at/ B G Outlier-Preserving Focus+Context Visualization in Parallel Coordinates D S H

Conclusions § Data abstraction based on density rep. § Focus+Context § § § Much clearer view for large data Screen-oriented and output-sensitive Interactive visualization of large data 35 § Data operations - outlier detection, clustering § Variable context precision § Outliers preserved Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Acknowledgements § § § § 36 K-Plus Vega grant 1/3083/06. AVL List Gmb. H - data Juergen Platzer Prof. Peter Filzmoser Harald Piringer Michael Wohlfahrt Matej Novotný http: //www. VRVis. at/ Outlier-Preserving Focus+Context Visualization in Parallel Coordinates

Thank you for your attention!