Categorical data is data that contains attributes with only a few different values. Examples for such data are sex, ethnicity, age group, etc. In many cases, there isn't even an inherent ordering of these values, and the differences between the values are not meaningful.

Even though categorical data is very common in real-world data sets, it has so far been mostly neglected, and treated as a special case of continuous data. Parallel Sets take a radically different approach. Based on the parallel axis alignment of parallel coordinates, the new method shows the number of values in each category, rather than the individual data points. This way, the typical problem of plotting many data values on the same point is avoided, and the display is much more useful. The image shows three dimensions of the Titanic data set: class (crew, first, second, third), sex (male, female), and survived (no, yes). By following the "ribbons" from one axis to the other, you can find out how each of the categories is subdivided, and what the survival rates were for different groups of people on the ship.

Parallel Sets also include a complex set of interactions, that are supported by meta data. Meta data describes the categories and the values in them, but can also contain hierarchies of the dimensions as well as the categories. This way, the user can interact with the data in the way he or she is already thinking about it, and does not have to adapt to the visualization too much. Hierarchical categories can be folded, yielding a more abstract view of the data, or unfolded, revealing more details. In addition, the user can create new dimensions for the existing ones, and thus define properties in more natural ways. Knowledge can be encoded in these definitions, and they can serve as the basis for further analysis.

Fabian Bendix, Robert Kosara, Helwig Hauser, Parallel Sets: Visual Analysis of Categorical Data. Proceedings of the 2005 IEEE Symposium on Information Visualization (InfoVis), pp. 133–140, 2005.

Robert Kosara, Fabian Bendix, Helwig Hauser, Parallel Sets: Interactive Exploration and Visual Analysis of Categorical Data, Transactions on Visualization and Computer Graphics (TVCG), vol. 12, no. 4, pp. 558–568, July/August 2006.

This work was done for Inte:Ligand, a company that provides services and develops software for in-silico drug discovery.

The protein (or macromolecule) is shown to enable the user to pick a particular binding site. An abstracted depiction is used, reducing several thousand atoms to just the backbone of the protein, which is shown as one long, winding ribbon. Different areas of interest (alpha-helices, beta-strands) are color-coded to make identification easier. In this case, there are two ligands in the protein, which are drawn into it, and surrounded by boxes. Upon clicking one of these boxes, the view zooms in to this particular binding site.

After zooming into a binding site, the user is presented with a close-up view of the ligand. Several rendering styles are available, which mostly adhere to the standard ways of depicting molecules in chemistry. Here, the ligand is shown using stick mode, while its environment (those parts of the protein within less than 5 Ångstroms) is rendered as lines. The user can interact with the molecule by rotating it, changing the render style for ligand, environment or individual atoms and bonds, adding or removing a pharmacophore, and moving parts of the molecule to or from the environment.

The pharmacophore can be drawn in addition to the molecule. The yellow spheres describe areas of the molecule that are lipophilic, while the green pointers show hydrogen acceptors. The pharmacophore describes a blueprint of the interactions between the ligand and the protein, which serves as a basis for finding other molecules with similar effects. The image shows the use of SDOF to align several possible ligands manually.

Gerhard Wolber, Robert Kosara, Pharmacophores from Macromolecular Complexes with LigandScout, in Langer, Hoffmann (eds), Pharmacophores and Pharmacophore Searches, pp. 131–150, Wiley, 2006.

Combined 2D/3D Scatterplots provide a way to display and interact with large datasets. Depth cues such as color, point size, and occlusion/halos are combined with a high frame rate to provide a good depth impression. Three 2D views provide the means for interaction, which is difficult in 3D. When zooming into the data set, histograms of the points outside the current viewpoint are drawn to give an impression of the surrounding data space. Histograms also provide feedback when brushing.

Harald Piringer, Robert Kosara, Helwig Hauser, Interactive Focus+Context Visualization with Linked 2D/3D Scatterplots. 2nd International Conference on Coordinated & Multiple Views in Exploratory Visualization (CMV), 2004.

Histograms are very useful for gaining an overview over large data. The data is reduced and abstracted, which makes it easier to see the overall structure, rather than every single data point. However, standard histograms do not work very well for data with a temporal dimension, like the results of computational fluid dynamics (CFD) simulations. TimeHistograms are an extension of the traditional histogram that make it possible to examine time-dependent ("unsteady") data sets. Three different modes were implemented that provide information about the time dimension, and allow the user to choose the amount of informmation displayed.

Robert Kosara, Fabian Bendix, Helwig Hauser, TimeHistograms for Large, Time-Dependent Data. Joint Eurographics - IEEE TCVG Symposium on Visualization (VisSym 2004), pp. 45-54, 2004

Asbru is the name of a language that is used to describe therapy plans in the Asgaard Project. The complexity of the language and the low familiarity of physicians with computer concepts made it necessary to develop a visualization and user interface to understand the concepts of the language, as well as author treatment protocols in it. AsbruView provded a simple, metaphor-based depiction of plans and their interactions. Two different views were designed, to show different aspects of the plans. In topological view, traffic signs and lights represent the different types of conditions (preconditions, complete, abort, suspend, etc.). The plans are put on a symbolic timeline, which only represents their temporal order, but not their extent.

In temporal view, the time annotation glyph is used to show detailed information about the temporal extent of plans or conditions. A time annotations consists of a reference point and four time points: the earliest starting shift (ESS), latest starting shift (LSS), earliest finshing shift (EFS) and latest finishing shift (LFS). The minimum and maximum durations (MinDu and MaxDu) can be set withing certain bounds, which are also shown in the visualization.

Temporal view shows the complex temporal extent of plans and conditions. Simple symbols are used to indicate the type of plan, which defines the possible temporal layout. Arrows represent any-order plans, while optional plans are shown with question marks.

Robert Kosara, Silvia Miksch, Metaphors of Movement: A Visualization and User Interface for Time-Oriented, Skeletal Plans. Artificial Intelligence in Medicine, Special Issue on Information Visualization in Medicine, pp. 111-131, Volume 22, Number 2, 2001.