Interactive Exploration of Multidimensional DataIs It Really That Common ?I can’t see it, I tell ya !Standard SolutionBut there are so many !Introducing Rank-by-featureHow does it work ?Ranking Criteria - 1DRanking Criteria - 2DPut A Demo Where Your Mouth Is !PowerPoint PresentationSlide 12Slide 13Slide 14Slide 15Questions/Critiques1Interactive Exploration of Multidimensional DataInteractive Exploration of Multidimensional DataBy:Sanket SinhaNitin MadnaniBy:Sanket SinhaNitin Madnani2Is It Really That Common ?Is It Really That Common ?You Bet:DemographicsEconomicsCensusMicroarray Gene ExpressionEngineeringPsychologyHealthYou Bet:DemographicsEconomicsCensusMicroarray Gene ExpressionEngineeringPsychologyHealth3I can’t see it, I tell ya !I can’t see it, I tell ya !Visualization challenges for >= 3D:Relationship comprehension is difficultDiscovering outliers, clusters and gaps is almost impossibleOrderly exploration is not possible with standard visualization systemsNavigation is cognitively onerous and disorienting (3D)Occlusion (3D)Visualization challenges for >= 3D:Relationship comprehension is difficultDiscovering outliers, clusters and gaps is almost impossibleOrderly exploration is not possible with standard visualization systemsNavigation is cognitively onerous and disorienting (3D)Occlusion (3D)4Standard SolutionStandard SolutionCan you say “Pro-jek-shun” ?Use lower dimensional projections of data:Can you say “Pro-jek-shun” ?Use lower dimensional projections of data:1D : Histograms2D : Scatterplots5But there are so many !But there are so many !For 13 dimensions (columns) : Number of histograms = 13Number of scatterplots = C(13,2) = 78Must examine a series of these to gain insightsUnsystematic == InefficientMust have order !For 13 dimensions (columns) : Number of histograms = 13Number of scatterplots = C(13,2) = 78Must examine a series of these to gain insightsUnsystematic == InefficientMust have order !6Introducing Rank-by-featureIntroducing Rank-by-featureAllows projections to be examined in an orderly fashionA powerful framework for interactive detection of:Inter-dimension relationshipsGapsOutliersPatternsAllows projections to be examined in an orderly fashionA powerful framework for interactive detection of:Inter-dimension relationshipsGapsOutliersPatterns7How does it work ?How does it work ?Framework defines ranking criteria for 1D & 2D projectionsUser selects criterion of interestAll projections are scored on the criterion and rankedUser examines projections in the order recommendedEureka* !!Framework defines ranking criteria for 1D & 2D projectionsUser selects criterion of interestAll projections are scored on the criterion and rankedUser examines projections in the order recommendedEureka* !!*Disclaimer: All users may not be able to make life-altering discoveries8Ranking Criteria - 1DRanking Criteria - 1DNormality: Indicative of how “Gaussian” the dataset isUniformity: How “uniform” is the dataset ?(How high is the entropy ?)Outliers: The number of potential outliers in the datasetGap: The size of the biggest gapUniqueness: Number of unique data pointsNormality: Indicative of how “Gaussian” the dataset isUniformity: How “uniform” is the dataset ?(How high is the entropy ?)Outliers: The number of potential outliers in the datasetGap: The size of the biggest gapUniqueness: Number of unique data points9Ranking Criteria - 2DRanking Criteria - 2DLinear Correlation: Pearson’s correlation coefficientLSE: Least Square Error from the optimal quadratic curve fitQuadracity: Quadratic coefficient from fitting curve equationUniformity: Joint entropyROI: Number of items in a Region Of InterestOutliers: Number of potential outliersLinear Correlation: Pearson’s correlation coefficientLSE: Least Square Error from the optimal quadratic curve fitQuadracity: Quadratic coefficient from fitting curve equationUniformity: Joint entropyROI: Number of items in a Region Of InterestOutliers: Number of potential outliers10Put A Demo Where Your Mouth Is !Put A Demo Where Your Mouth Is !11HCE OverviewHCE Overview12The Input Dialog BoxThe Input Dialog BoxPerform Filtering & NormalizationPerform Filtering & Normalization13Histogram OrderingHistogram Ordering14Scatterplot OrderingScatterplot Ordering15Tabular View of DataTabular View of DataSelect specific data records and annotate if neededSelect specific data records and annotate if needed16Questions/CritiquesQuestions/CritiquesWhat does “outlierness” mean?Cannot identify datapoints in histogram or scatterplot browser without switching to table viewEspecially in ROIHow to intuitively interpret:Outliers in 2DLSEQuadracityWhat does “outlierness” mean?Cannot identify datapoints in histogram or scatterplot browser without switching to table viewEspecially in ROIHow to intuitively interpret:Outliers in
View Full Document