Connor Gramazio

Colorgorical: Creating discriminable and preferable color palettes for information visualization

We present an evaluation of Colorgorical, a web-based tool for creating discriminable and aesthetically preferable categorical color palettes. Colorgorical uses iterative semi-random sampling to pick colors from CIELAB space based on user-defined discriminability and preference importances. Colors are selected by assigning each a weighted sum score that applies the user-defined importances to Perceptual Distance, Name Difference, Name Uniqueness, and Pair Preference scoring functions, which compare a potential sample to already-picked palette colors. After, a color is added to the palette by randomly sampling from the highest scoring palettes. Users can also specify hue ranges or build off their own starting palettes. This procedure differs from previous approaches that do not allow customization (e.g., pre-made ColorBrewer palettes) or do not consider visualization design constraints (e.g., Adobe Color and ACE). In a Palette Score Evaluation, we verified that each scoring function measured different color information. Experiment 1 demonstrated that slider manipulation generates palettes that are consistent with the expected balance of discriminability and aesthetic preference for 3-, 5-, and 8-color palettes, and also shows that the number of colors may change the effectiveness of pair-based discriminability and preference scores. For instance, if the Pair Preference slider were upweighted, users would judge the palettes as more preferable on average. Experiment 2 compared Colorgorical palettes to benchmark palettes (ColorBrewer, Microsoft, Tableau, Random). Colorgorical palettes are as discriminable and are at least as preferable or more preferable than the alternative palette sets. In sum, Colorgorical allows users to make customized color palettes that are, on average, as effective as current industry standards by balancing the importance of discriminability and aesthetic preference.

MAGI: visualization and collaborative annotation of genomic aberrations

1Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany. 2Janelia Research Campus, Howard Hughes Medical Institute, Ashburn, Virginia, USA. 3Department of Anatomy and Structural Biology, Albert Einstein College of Medicine, Bronx, New York, USA. 4Gruss Lipper Biophotonics Center, Albert Einstein College of Medicine, Bronx, New York, USA. e-mail: pietzsch@mpi-cbg.de or tomancak@mpi-cbg.de

The relation between visualization size, grouping, and user performance

In this paper we make the following contributions: (1) we describe how the grouping, quantity, and size of visual marks affects search time based on the results from two experiments; (2) we report how search performance relates to self-reported difficulty in finding the target for different display types; and (3) we present design guidelines based on our findings to facilitate the design of effective visualizations. Both Experiment 1 and 2 asked participants to search for a unique target in colored visualizations to test how the grouping, quantity, and size of marks affects user performance. In Experiment 1, the target square was embedded in a grid of squares and in Experiment 2 the target was a point in a scatterplot. Search performance was faster when colors were spatially grouped than when they were randomly arranged. The quantity of marks had little effect on search time for grouped displays (“pop-out”), but increasing the quantity of marks slowed reaction time for random displays. Regardless of color layout (grouped vs. random), response times were slowest for the smallest mark size and decreased as mark size increased to a point, after which response times plateaued. In addition to these two experiments we also include potential application areas, as well as results from a small case study where we report preliminary findings that size may affect how users infer how visualizations should be used. We conclude with a list of design guidelines that focus on how to best create visualizations based on grouping, quantity, and size of visual marks.

Heapviz: Interactive heap visualization for program understanding and debugging

Understanding the data structures in a program is crucial to understanding how the program works, or why it does not work. Inspecting the code that implements the data structures, however, is an arduous task and often fails to yield insights into the global organization of a program’s data. Inspecting the actual contents of the heap solves these problems but presents a significant challenge of its own: finding an effective way to present the enormous number of objects it contains. In this paper we present Heapviz, a tool for visualizing and exploring snapshots of the heap obtained from a running Java program. Unlike existing tools, such as traditional debuggers, Heapviz presents a global view of the program state as a graph, together with powerful interactive capabilities for navigating it. Our tool employs several key techniques that help manage the scale of the data. First, we reduce the size and complexity of the graph by using algorithms inspired by static shape analysis to aggregate the nodes that make up a data structure. Second, we implement a powerful visualization component whose interactive interface provides extensive support for exploring the graph. The user can search for objects based on type, connectivity, and field values; group objects; and color or hide and show each group. The user may also inspect individual objects to see their field values and neighbors in the graph. These interactive abilities help the user manage the complexity of these huge graphs. By applying Heapviz to both constructed and real-world examples, we show that it provides programmers with a powerful and intuitive tool for exploring program behavior.

Exploring hierarchical visualization designs using phylogenetic trees

Ongoing research on information visualization has produced an ever-increasing number of visualization designs. Despite this activity, limited progress has been made in categorizing this large number of information visualizations. This makes understanding their common design features challenging, and obscures the yet unexplored areas of novel designs. With this work, we provide categorization from an evolutionary perspective, leveraging a computational model to represent evolutionary processes, the phylogenetic tree. The result - a phylogenetic tree of a design corpus of hierarchical visualizations - enables better understanding of the various design features of hierarchical information visualizations, and further illuminates the space in which the visualizations lie, through support for interactive clustering and novel design suggestions. We demonstrate these benefits with our software system, where a corpus of two-dimensional hierarchical visualization designs is constructed into a phylogenetic tree. This software system supports visual interactive clustering and suggesting for novel designs; the latter capacity is also demonstrated via collaboration with an artist who sketched new designs using our system.

Which color means more? An investigation of color-quantity mapping in data visualization

Colormaps representing quantities are a standard tool of data visualization (e.g., in correlation matrices, brain activation diagrams, and topographic maps). We investigated peoples implicit intuitions about how the extremes of color scales (e.g., dark/light) map onto represented quantities (e.g., large/small, high/low), and whether they are modulated by contrast with the background. Participants were shown fictitious data matrices in which columns represented time, rows represented alien species, and cell color represented how often each species was spotted during each time window. Each participant was shown one colormap and indicated whether there were more animals early or late. Early/late differences were clearly present, but no legend indicated how to interpret the colors. We expected higher-contrast would correspond to larger quantities. In Experiment 1 we tested dark-red/light-orange and dark-blue/light-cyan colormaps on black and white backgrounds. Almost everyone (89%-95%) inferred that darker colors represented larger quantities. Surprisingly, the effect was not modulated by contrast with the background, perhaps because all the colors were relatively high-contrast. In Experiment 2, we used the same colormaps but the background colors were extensions of the scale endpoints (e.g., darker blue or lighter cyan for the dark-blue/light-cyan colormap) so that one endpoint was always low-contrast. We also tested a gray-scale colormap on white and black backgrounds. Against light backgrounds, 90%-98% showed the dark-is-more bias. Against dark backgrounds, significantly fewer participants showed the same bias (53%-74%), but the pattern did not reverse. Therefore, participants have a strong dark-is-more bias, which is diluted, but not reversed, when dark colors are low-contrast. In a systematic survey of colormaps in published visualizations (e.g., in Nature Neuroscience) we found that many violate the dark-is-more bias. Using empirically validated natural intuitions for color-concept associations will help make complex datasets easier to understand. Meeting abstract presented at VSS 2015.

Optimizing an SPT-tree for visual analytics

Despite the extensive work done in the scientific visualization community on the creation and optimization of spatial data structures, there has been little adaptation of these structures in visual analytics and information visualization. In this work we present how we modify a space-partioning time (SPT) tree - a structure normally used in direct-volume rendering - for geospatial-temporal visualizations. We also present optimization techniques to improve the traversal speed of our structure through locational codes and bitwise comparisons. Finally, we present the results of an experiment that quantitatively evaluates our modified SPT tree with and without our optimizations. Our results indicate that retrieval was nearly three times faster when using our optimizations, and are consistent across multiple trials. Our finding could have implications for performance in using our modified SPT tree in large-scale geospatial temporal visual analytics software.