Discussion questions from CS 7390 lecture

Date: 1/29/96

Presented by: John Stasko

Topic: Algorithm Animation

• The Brown paper brings up (again) the dividing line between Program Visualization and Visual Programming. Has anyone explored the areas where they overlap? And would it be worth it? (e.g., Is it easier to make algorithm animations of routines "written" in a Visual Programming Language? Or can you start bay creating an animation and have enough to be able to compile the algorithm?)
• It sounds like Cox & Roman use Prolog (or at least something similar) for part of their animation system. Did they? Is there something else that could be used to "solve" their rules?
• Although Cox and Roman's paper illustrates the use of Pavane to animate an algorithm, and also refers to the use of Balsa and Tango, no mention is made of the effectiveness of any of them in conveying how the algorithm works to someone other than the animator. Are algorithm animations effective, especially to someone not familiar with the algorithm being animated?
• How does the path-transition paradigm in Tango map program actions to animation actions? How does Tango know which program actions are significant?
• A general question about taxonomies: Even if they are sometimes large, and well thought about, there seems to be little information in it about how the authors derive them. From direct observation? In those studies, is there enough attention and centering on the users, their needs and visualization goals, and whether they reach them?
• About algorithm animation: A system is descrived to try to be automatic algorithm display. How far can it go? Can you display something (beyond simple structures) without underestanding at least part of its meaning before?
• Does Cox's classification of algorithm animation encompass Brown's taxonomy which is highly focused on?
• In the path-transition paradigm of Tango, there may be many significant paths. How can it decide which path to transition?
• In the Wash. U paper, the brunt of the discussion was on how visualizations of algorithms can be classified by their levels of abstraction. It treats the visualization as a mapping from the domain (scope of information extracted) to the range (graphical techniques used). Can the scope of information extracted be generalized enough that it can be used to characterize algorithm animations or is it too specific to each individual algorithm?
• In Brown's paper, there is a taxnomy given for algorithm animations: content, persistence, and transformation. Does this classification cover all aspects or should it be augmented with the general visualization taxonomy we constructed in class: interface, effectiveness, and method?
• Is there any way to create a redundant encoding for animation? (so that one single frame would contain information suggesting that it was part of an animation. Example: use dotted lines or shadow paths)
• Given that algorithm animation can't be done automatically, what are its uses? Pedagogy, documentation, videos?
• What are the important differences between creating a taxonomy for general software visualization and creating one for algorithm animation?
• Do the "discrete transformations" discussed in Brown's paper still meet Stasko's taxonomy definition of animation?
• Automated algorithm animation apparently requires the semantics of the algorithm to be explicit in the program. Why not develop a new language to do that? This new language requires strong typing and choosing of meaningful function calls. Object-oriented language is working toward that direction. This new languaage will not only aid algorithm animation, it will also have significant impact on software engineering, since the code is in correspondence with the meaning of the algorithm now.
• Different levels of abstraction in algorithm animation correspond to different levels of abstraction in the algorithm to make the correspondence explicit in the code and its animations is very likely a job that can only be done by the programmer, because no machine can do any significant abstraction nowadays.
• In Roman & Cox, "Analytical" and "Explanatory" representations are each classified as one of the five distinct levels of abstraction. To what degree are they actually forms of synthesized representations?
• Brown, in his discussion of perspectives on algorithm animation, lays out a model for characterizing algorithm animations along three dimensions: Transformation, content, and persistence. His content dimension only spans from direct representations to synthetic. Could we simply replace this dimension with Cox/Roman's levels of abstraction? Would this still give us an orthogonal basis for classifying animations, or could it potentially cause conflicts? Why?
• What about specifying program animation events in some meta-language that calls callbacks that are object methods? This takes advantage of encapsulation of OO languages.
• How can the user affect the speed/direction of an animation? Should there be a standard control panel for speed? Should it be content specific?