Multiscale Visualization Using Data Cubes Chris Stolte Diane

Multiscale Visualization Using Data Cubes Chris Stolte, Diane Tang, Pat Hanrahan Stanford University Information Visualization October 2002 Boston, MA

Motivation Large multidimensional databases have become very common § Need techniques for exploration and analysis § “Overview first, zoom and filter, then detailson-demand” §

Multiscale Visualization § Visual representation changes as user pans and zooms • Overview, lots of data highly abstracted • Zoom, data density decreases detailed information shown § Visual and data abstraction • Visual abstraction different representation/same data • Data abstraction transformations to reduce data set size

Existing Multiscale Visualizations Cartography § Multiscale information visualization § • • § Pad++: alternate desktops Data. Splash Xmdv. Tool ADVIZOR Main limitations: • One zoom path • Primarily visual abstraction

Contributions § Multiscale visualization with both visual and data abstraction using generalized mechanisms: • Data Abstraction Data Cubes • Visual Abstraction Polaris § “Design Patterns”

Data Cubes

Data Warehouses § § Store data for analysis (OLAP) Fact table contains measures categorized by dimensions: Fact table State Ordinal fields Month (categorical dimensions) Product Name Profit Sales Quantitative fields Payroll (measures) Marketing Inventory Margin. . . Coffee chain (courtesy Visual Insights)

Hierarchical Structure § § Data warehouses are very large—need to summarize Add hierarchical structure to warehouse Dimension tables Location Market State Products Product Type Product Name Fact table State Month Product Name Profit Sales Payroll Marketing Inventory Margin. . . Time Year Quarter Month

Hierarchical Dimensions § Each dimension table describes a tree § Each level describes a level-of-detail § Meaningful basis for aggregation § Create summaries of fact table for each level-of-detail as Data Cubes Time Year Quarter Month

Data Cube § § Create cube for each level-of-detail combination Summary of fact table Cube for (Quarter, Product Type, Market) Each cell aggregates all measures for those dimensions Each cube axis corresponds to a dimension in the relation at a level-of-detail

Hierarchies & Data Cubes § Hierarchies define a lattice of cubes: Least detailed Each cube is defined by a level-of-detail in each dimension Data abstraction Most detailed

Projecting Data Cubes § Can further abstract a cube by “projection” Data abstraction

Data Cube Summary § Industry standard for storing analytic data § Provide summaries of data at meaningful levels of detail § To perform data abstraction: • Design a hierarchical schema • Choose a cube in the lattice of cubes • Project to relevant dimensions § Identifying a projection corresponds to specifying the desired data abstraction

Polaris

Exploring Data Cubes using Polaris § Polaris is: • A UI for exploration, analysis of data warehouses • A formal language for specifying queries & visualizations • An interpreter for compiling specification into queries/drawing commands § Demo!

Polaris Formalism § Visualization described using visual specifications that define: • Table configuration (algebra) • Type of graphic in each pane • Encoding of data as visual properties of marks (encoding system) • Data transformations and queries § Each specification corresponds to a projection of the data cube

Path of Exploration § Can think of an analysis as path of specifications

Path of Exploration Visual abstraction

Path of Exploration This is a multiscale visualization! Data abstraction

Graphical Notation

Graphical Notation: Templates Instance Template

Specifying Multiscale Visualizations Specify multiscale visualization using a graph of Polaris specifications: a Zoom Graph § Paper describes how to implement using Polaris § Zooming Polaris Specification Possible zoom

Specifying Multiscale Visualizations § Can specify a zooming pattern by using templates

Specifying Multiscale Visualizations § Independent zooming on different dimensions is described as a graph y-axis zoom x-axis zoom

Design Patterns

Design Patterns § Zoom graphs simplify specifying and implementing multiscale visualizations • Design is still very hard § “Design patterns” (a la Gamma et al. ) • Capture zoom structures that have been used effectively & reuse in new designs • We present four such patterns • Formal way to discuss multiscale visualization

Thematic Maps

Chart Stacks

Matrices

Dependent QQ Plots

Summary

Summary § Multiscale visualization with both visual and data abstraction using generalized mechanisms: • Data Abstraction Data Cubes • Visual Abstraction Polaris § “Zoom Graphs” for specifying and implementing multiscale visualizations § “Design Patterns”

Future Work Designing new patterns § Transitions between levels-of-detail § • Communicate parent-child relationships • Non-uniform branching • Animation/dissolve/fade? § Data management • Prefetching and caching of large data sets