Selective Dynamic Manipulation of Visualizations Chuah, Roth, Mattis, Kolojejchick

Motivation Need 3D techniques for interactive visualizations of multidimensional data. We want: –Selective: A high degree of user control –Dynamic: Interactions all occur in real time, with animation –Manipulation: Users can directly move and transform objects in the visualization Author’s system is called SDM

Barriers Many data sets have too much information to be on screen at once Much clutter and occlusion (hidden data) in dense sets of data Difficult to give a sense of scale: some objects may be completely dwarfed by others (green objects in Fig 1)

More Barriers... Must be able to classify data into sets and save those classifications Must be able to compare quantities which are not near each other (difficult to compare heights, for example, if they are at different distances from the user [Fig 3]) Authors believe SDM deals with these issues

Sample Data Crisis relief network –Supply centers are cylinders –routes between them are dark lines on the floor –shelters where supplies are needed are rectangular bars –heights of cylinders and bars indicate supplies available or needed –Arranged in a network, like Becker paper

SDM components Object centered selection –the selected set is made up of objects instead of a spatial area –can click on desired objects or use our old friend the constraint slider –when you create them, you can save and name them

More SDM components Dynamic operations –The user uses a “physical” handle to manipulate the data (Fig. 4) –Attach a handle to an object, and push or pull on it: causes the object, or a set of objects, to grow, shrink or move –can control one or more parameters with single handle

Constraints Context persistence –SDM maintains a relation between the set being manipulated and the original set. Set wide operations –if you can move or scale one object in a focus set, you can move or scale any.

Feedback Techniques SDM must clearly identify the selected set –so we know what objects will change if we take an action SDM must maintain scene context –if we change something, a “shell” of the original value is left in its original place.

More Feedback Techniques Maintain Temporal Continuity –They use animation to allow the user to see what has happened without having to think too hard about it Maintain relationship between selected set and environment –Keep a scale of the differences on screen, for example Allow objects easily to be returned to their original positions

How Do We Use It? You can tell selected set apart by color or width You can view occluded objects by –elevating them (Fig 9) - lose context –making all other objects invisible - lose context –making all other objects of height 0 (Fig 10) - lose context –make all other objects very thin (Fig 11) - still partially hidden –make other objects transparent

Favorite Sentence … the “physics” provided by SDM is not limited to real world manipulations; users can also elevate, compress, and perform other operations upon objects that wouldn’t be possible with actual physical models.

How do we use it? (cont) For different data sets, can use different scales –This is so that data sets with much greater or lesser values do not dominate Can interactively make and visualize new classes of data –This is a lot better than having to update the entire database first

How Do We Use It? To solve the problem of comparing things at different distances, sets of data can be brought to the front and compared in two dimensions (Fig 13)

Strengths Enables a more precise, quantitative comparison between objects –preserves relationships between focus objects and rest of data –scaling is kept correct –distortions and occlusions of 3D are overcome Also, it is pretty cool

Weaknesses Can only view limited part of the data set: the rest may be “in the distance” (possibly add rotation) Can still get occlusion problem if focus set is dense Does not address multimedia, UI, how to decide on representation?

Continuing Efforts Sage research project –SDM’s “physicalization” of the abstract space is combined with automated visualization tools, multimedia and UI stuff to create an entire system

Externalizing Abstract Mathematical Models Tweedie, Spence, Dawkes, Su

The Problem Mathematical models are important in many domains They are often quite complex, not having an obvious physical visualization –an example of an obvious one would be a flow model might into a network or a pipe How can we visualize them?

The Solution Interactive Visualization Artifacts (IVAs) –Instead of visualizing the raw data, we visualize precalculated data as 2 kinds of data a description of the physical nature of an artifact, called parameters a description of the results we can expect from an artifact, called performance criteria We develop different IVAs to handle any given problem - we describe 2

Our Example The Light Bulb –design parameters: filament width, filament material –performance criteria: cost, brightness, lifetime But there are problems

The Problems with Light Bulbs We need to create a light bulb given the performance data: but there is no way to get the parameters given the performance data (except trial and error - ugh!) Changes in manufacturing mean that any set of parameters can only be guaranteed to be in a range of values - but not exact values

More Problems with Light Bulbs Often, you also have to maximize some other objective, like manufacturing yield.

IVA One: The Influence Explorer We precalculate the data and display histograms based on it [Fig 6] Each bulb design is represented once for each parameter and criterion: the design goes in the appropriate bin The upper and lower limits on the sliders can represent the desired limits (Red passes all performance requirements, and black to white indicates it has failed some)

The Influence Explorer If we also want to chart performance and parameters, we can do so as in Fig 7 Red is correct for all Blue means it fails some performance requirement (thus will reduce yield, but can still be made) Black, gray or white means it has failed one or more performance and/or parameter requirement

Influence Explorer This color coding shows how altering the criteria will help Keep playing around until yield (which is computed and shown) is high

IVA 2: The Prosection Matrix Provides a scatterplot for each possible pair of parameters [Fig 13, 14] –This is a 2D PROjection of a SECTION of n- dimensional parameter space.

Prosection Matrix, cont Values are chosen at random to be projected on the scatterplot from the performance requirements given –Can adjust the sliders to determine the acceptable performance requirements –Place a bounding box in the section to determine ranges of parameters

Strengths Reasonably effectively maps multivariable data into 2 dimensions Can transform a complicated problem into a much simpler one Influence Explorer is partially analogous to parallel coordinates –can use intuitions from that representation

Weaknesses Some of these problems can reasonably be automated (hill climbing algorithms, etc) Prosection matrix makes you reduce problems to pairs of criteria –counter-intuitive projection representation May not effectively handle large numbers of variables –(n 2 - 3n + 2) /2 prosection matrices is a lot.