610 likes | 833 Views
Lecture 20: High Dimensional Visualization. April 23 , 2013 COMP 150-2 Visualization. Recall…. We have examined the relationship between data type and data dimensionality to the appropriateness of the visualization. Recall…. Some visualizations that we implemented in class…. Recall….
E N D
Lecture 20:High Dimensional Visualization April 23, 2013 COMP 150-2Visualization
Recall… • We have examined the relationship between data type and data dimensionality to the appropriateness of the visualization
Recall… • Some visualizations that we implemented in class…
Recall… • They might be higher in dimensionality, but they are not necessarily generalizable. • Force-directed layout: specific to relational data • ThemeRiver: specific to temporal data of discrete channels • Treemap: specific to hierarchical data • etc.
Multi-dimensional Data Visualization • Data with 4 or more dimensions are considered to be multi-dimensional. • If a visualization technique is truly generalizable, it should be able to handle 10-20 dimensions.
Fundamental Challenge • Your monitor (screen / canvas) is inherently 2 dimensional… • Trying to visualize data with dimensionality > 2, by definition, requires dimension reduction or dimension projection. • This means finding some way to map each dimension onto a 2D plane • A very challenging problem….
Most Common Example • Spreadsheet • Maps dimensions to columns • Data items into rows • Projects the high dimensions into small rectangular regions (column)
Other Examples We’ve Encountered… http://hesketh.com/schampeo/projects/Faces/chernoff.html
Other Examples • Tufte’s Small Multiples
Dynamic Home Finder • (Williamson, Shneiderman, 1992)
Table Lens • Rao, Card (1994)
Table Lens • Video: http://www.open-video.org/details.php?videoid=8304
FOCUS • Flips the Table Lens • Columns are data items • Rows are attributes • Emphasis on the dimensions in the data (not individual data items)
FOCUS • Sort on any attribute along the row • Querying along the rows (attributes)
Parallel Coordinates • Different dimensions have different ranges • Need to normalize the dimension (to 0-1) • Need coloring…
Dimension Orientation DasguptaInfoVis 2010
Density Reduction Johansson EuroVis 2008
Angular Brushing http://www.vrvis.at/via/research/ang-brush/parvis4.mov Hauser InfoVis 02
Lots of Extensions… Drosophila Gene Expression Data Exploration and Visualization Lawrence Berkeley National Lab
Lots of Extensions… Villaveces, UNC / Renci Streit, Cytometry Part A, 2006 Fanea et al. InfoVis 05
VisLink • Connects a dot across different visualizations • Similar in concept to Parallel Coordinates, but certainly different… • http://faculty.uoit.ca/collins/research/VisLink/flash/index.html Collins et al. InfoVis 2007
Star Plots • Space out the dimensions across a circle with equal angles • Similar to Parallel Coordinates, but “wraps around”
Star Coordinates • Similar idea as star plot, but only plots a single dot (as opposed to a line or an area) • Presented by Georges Grinstein when he visited earlier this semester
Using Glyphs Chan, “A Survey on Multivariate Data Visualization”, 2006 Kindlmann, Vis, 2006
Categorical Data • Most examples shown are based on quantitative data • What about categorical data? • Nominal or Ordinal • Students • Gender: M/F • Eye color: Brown, Blue, Green, Hazel • Hair color: Black, Red, Brown, Blonde, Gray • Country: USA, Canada, China, Japan, India, Italy • Shown in Parallel Coordinates?
Mosaic Plot • Concept very similar to Treemap • Divide a square into sub-squares based on the attributes • Impose an ordering of the dimensions
Mosaic Plot • Example: Titanic Survivor • http://www.scribd.com/doc/2919000/Mosaic-plot-of-the-Titanic-survivors
Attribute Explorer • Tweedie CHI 1994 • Merges three concepts: • Tufte’s “Small Multiples” • Histograms • Coordinated visualization
Attribute Explorer http://www.open-video.org/details.php?videoid=8162