Visual flying rules

Share this on social media:

Felix Grant dons his khaki trousers and becomes an intrepid data explore.

In my younger days I flew gliders and, until an unfortunate incident that wrecked a hangar and flattened a perimeter fence, small spotter helicopters. Instruments were essential, but most of the useful work (and all of the fun) came under 'visual flying rules'. Rock climbing, walking and scuba diving may seem unconnected, but draw on the same set of hardwired, terrain-related psychomotor skills and perceptions. The whole terrain-related apparatus, part of our evolutionary heritage, is important to understanding any complex data input - physical or conceptual.

Anyone with a particularly strong aptitude in one direction is bound to be a bit odd, and in most mathematicians or statisticians the oddness comes out as a slightly autistic tendency to see patterns, structures or platonic forms where real people see messy existence. We are not usually very good at explaining our view of the world, but we almost invariably try to do it through visual metaphor. When my maths teacher asked how I'd gotten the solutions to a cubic roots problem I muttered that I'd 'watched them grow', which wasn't the answer she wanted. Poets and novelists do it better: Marcus Potter, the brilliant but fragile mathematician of A S Byatt's Frederica Quartet, describes in the final novel[1] how he would ' mathematical problems by ... seeing a garden ... I used to release the problem into the garden and ... see the answer'.

Not all of us are so explicitly or primarily visual as Marcus Potter, but psychovisual interpretation is the most common exploratory device - the intuitive point of access to structures which are only later formalised - and, equally important, the primary means of transmitting complex information onward. Marcus' illustration is one instance of a more general class of terrain metaphors and, by extension, the three-dimensional metaphors of swimming or flight. All involve envisioned transit and manoeuvre through the problem space. In almost every case, visualisation is a vital precursor to data analysis, the prospecting or initial pathfinder reconnaissance stage. It doesn't solve problems, but it provides essential intelligence on the routes, spaces, relationships and fracture lines within a data set, which offer most promise of analytical yield. And, after the analysis is done, a picture is worth a thousand words in bringing back to those at home what we have seen on our travels.

Nowadays, with computerised plotting in fast GUI-driven environments, the underlying forms of such visual tropes don't have to be described or even imagined: their essence can be generated in ever increasing subtlety and sophistication from the comfort and safety of a armchair.

All this has been brought on by arrival on my desk of two new software packages. Both are from established sources; one is an upgrade to a well-known product, the other is a new product from an existing stable. Each represents a different point on the spectrum of approaches and user requirements.

Getting the right tool for the context is crucial: too much complexity in one circumstance can be as bad as too little power in another. Sigmaplot 10 and Voxler are good exemplars of different approaches.

SigmaPlot has been around longer than I have been using computerised technical graphics packages, and has changed home twice in its history. The earliest incarnation of it on which I can lay my hands is a DOS version, release numbered at 3·0, with 1993 file stamps, though a 1992 review in the journal Physical Therapy[2] refers to version 4·1 (still a DOS program, but boasting Windows 3 compatibility). Either way, in those years it was still in the hands of its original publisher, California-based Jandel Scientific. In 1996, to a mixture of excitement and doom-mongering among SigmaPlot's user base, Jandel were taken over by SPSS - where development was, for the most part, incremental. A few years on, SPSS refocused on core concerns and a tranche of scientific software including the Sigma range moved over in three phases to Systat Software International (SSI), a subsidiary of Bangalore-based Cranes Software. Cranes was a recent start-up, but has established a meteoric record of successful acquisition and development across a surprising range of technologies in a series of bold, risky, but successful gambles. Each phase of the acquisition was bigger than the acquiring company, yet the move pumped new blood into the acquired products: in the case of SigmaPlot, two serious upgrades (three if you count one released during the transition period) and a trebled market. Which brings us to release 10.0.

Voxler's story is a different one. Golden Software of Denver has been in continuous business for 23 years, evolving and retaining its product line in a measured way. The focus of that line is data visualisation of various kinds, all with clear links to geology and mining, but (apart from one exception, Strater, a well- and borehole-plotting program) with fully generic application scope. The flagship, from a general scientific data exploration point of view, has for a long time been Surfer, a surface modeller which, even at release 6 a decade ago, was described in one comparative survey as 'eye popping'[3] (it is currently in a mature release 8). What Surfer has so long done for surfaces is now being brought to volumetrics by new addition Voxler.

The philosophies and target markets of the two products are very different, with only a little overlap, and to use both would be perfectly logical for a user with a wide spread of visualisation needs. For the purposes of getting to know them, however, I applied both to one study in which they offered complementary contributions. The study concerned is a commercial one still in progress, but in essence it involves the impact of new technology on efficient use of an expensive strategic resource and quality of a key output. It's a problem easily represented as a multidimensional space metaphor; either package can equally easily manifest that space fully in my required five dimensions (three digitally on an x, y, z Cartesian frame, another two expressively using symbol size and colour), and Voxler will take more than that in its stride, though the illustrative examples here concentrate on enhanced visualisation of three at a time.

Also very different are the operational interfaces, though they share a willingness to follow function rather than fashion. That's not a coded implication that they are hard to use; far from it. Each is logical, and intuitive too, but the user is expected to intelligently recognise that the controls are focused on what is being done and not on preconceptions brought from elsewhere.

In the case of Voxler, the usual familiar visualisation environment is replaced with a CASE-style drag-and-drop block assembly interface. Modules (objects representing data, preparation steps such as grid generation or transforms and other computational or preparation processes, and aspects of graphic output from plot type to details of mark-up) are dragged or double-clicked from a tree in the far left 'library' pane to a 'network' pane alongside it. The output resulting from a completed network appears, and can be manipulated, in the main right-hand pane. (All of these screen panes can be moved from the default positions I've described to any convenient alternative, docked in various ways, and even made to autohide or hide reduce to tabs. It's a very flexible environment.)

Each module can either be dragged into place manually, its input and/or output nodes then being manually connected to those of other modules, or there is a semi-automated alternative. Each newly-added module (starting with the loaded data) is highlighted, and the visible contents of the module tree change to reflect only those options appropriate to it. Double-clicking one of those options immediately places it in the network, ready connected to the one before, and highlighting it in turn ready for the next decision. Modules have a red, amber, green 'traffic light' spot showing how ready it is for use.

Once the network is assembled and the visualisation generated, navigating through it is a matter of mouse (all three buttons, plus Ctrl and Shift variants, can be assigned to motion types as preferred), trackball or joystick control. The whole thing works beautifully, and I was fully acclimatised within half an hour - it would have been quicker still, if I'd had the humility to glance briefly at the 'getting started' instructions.

SigmaPlot's interface was described by one reviewer in the early 90s as 'quirky', but it has stood the test of time, evolving, but not fundamentally changing, and demonstrating its fitness for purpose. The difference from most graphing programs is that SigmaPlot borrows from some other genres the idea of wraparound tool bars: by default, two conventional ones at the top of the screen, two bulkier ones at the left and one at the right. Within that frame are a left-hand object tree and a right-hand visualisation output pane. The initial choice (after loading data, which is done from the file menu in the conventional Windows way) is of a basic chart type - one bar offering 2D options, the other 3D. Clicking one of these opens a pop-up option gallery with subtypes, and a wizard then guides subsequent choices through to a default output which can then be fine-tuned.

That fine-tuning is both a strength and the main SigmaPlot learning curve. You can fiddle to your heart's content with most aspects of the output, and this release brings new subtleties to further expand the range of choices to be made. New axes or axis options (including Weibull, Arrhenius and other reciprocal scale types), finer resolution in object selection, and so on, expand both the opportunities for control and the size of the forest to be navigated.

On the other hand, enhanced assistance in navigating that forest is also now available. At one end of the scale are small things like the stress-busting facility to 'click through' one object to another which lies behind it, or automatic generation of 95 per cent confidence bands when curve-fitting. At the other is the submission assistant, which is analogous to the stored publication styles in the likes of EndNote or Nota Bene's Ibid. The assistant's templates store specific requirements for different publications and check exports against them and can be modified, created, extended, to tailor the 20 or so provided examples to your own needs. Where several members of a team are collaborating on a notebook, SigmaPlot now maintains audit trails with specific user identification. Tutorials for first-timers, slicker data import, and other small touches all help to ease the load. There is also a macro-based pharmacological menu, built from the automation language, which makes life easier if you are working in that area (in which case you may well also be interested in the new ROC curves and associated tools).

More information is now stored by SigmaPlot from session to session, too, which reduces pressure while giving productivity a boost. Transforms (of which there are 12 new ones this time around) are stored in notebooks and associated with specific data files. Access to previously used documents, or their reuse as new templates, is smoother and easier. This may be as good a place as any to mention that start-up from cold now takes significantly longer than in previous versions, and for much of the extra time there is no visual sign on-screen that anything is happening, so it's easy to make the mistake of thinking that the program has frozen. You'll soon get into the habit of doing something else for a minute or two, though, and leaving the SigmaPlot window minimised rather than shutting it down between sessions.

Curve-fitting (explicit or implicit) is an important part of what these products do. On the explicit front, alongside the regression wizard (which has itself acquired piecewise-linear models in two to five segments, with automatic parameter estimation), SigmaPlot sports a new iterative curve fitter. Up to 25 coefficients are optimised across a maximum of 10 independent variables, with user control over 25 parameter constraints and the option to use reduced chi-squared weighting. Implicitly, the generation of grid lattices is done automatically on the fly, which makes surface fits fast and hassle free. Voxler's lattices (for isosurfaces, for instance) are performed by a gridder invoked from the library - trading away slickness for closer user-control.

SigmaPlot is page oriented, with a default assumption that your work will realise its destiny on paper, though it is equally capable of producing electronic output of various kinds. Some of the developments in this release, such as the streamlined automatic positioning and adjustment of graph groups, look specifically in that direction. Voxler, by contrast, is built primarily for onscreen explorations, though again it has all the necessary provision for printing (including a multi-page option for printing large output over several sheets with photographic quality). More important is the form and nature of that output - as with complexity, it's important to get the right tool for the job.

SigmaPlot will, for instance, produce a polar graph (with angle measured in either direction), or a ternary plot; Voxler will not. Voxler, on the other hand, will show you a view through (or even, if you zoom in enough, from within) a complex data structure with vector flows and particular planes or regions; SigmaPlot will not. All of this, of course, reflects the different intent within the two products; some things are appropriate in one context, not in another.

Voxler provides the gamut of stage lighting and atmospheric fog to sustain its communicative sensory illusions; SigmaPlot has substantially expanded its repertoire of line types and texture range to enhance discrimination. Combination of projected contours with 3D mesh plots (new in this SigmaPlot release), on the other hand, is a technique shared by both, but differently implemented.

Time to fall back on my opening metaphors for a quick summary. Both of these packages give powerful insight and communication tools to the intrepid professional data explorer, with choice between them depending on where you are going and what you plan to do when you get there. SigmaPlot will give you the high-quality Ordnance Survey maps to guide you as you hike across or clamber over your data, showing every detail of the terrain at any resolution you require; it will also give you the controls of a light aircraft in which to overfly that terrain, examining it from every possible bird's eye view.Voxler, instead, will provide you with the specialised skills to fly (or swim) in and out of the structures and spaces within your data. Either way, it's a thrilling ride.


1. Byatt, A.S., A Whistling Woman. 2003, London: Vintage (Random House). 2. Francis, K., “Sigmaplot 4.1. (Evaluation)”. Physical Therapy, 1992(72.n4): p. 1. 3. Simon, B., Data on display: Windows graphing programs. Desktop Engineering, 1996(1): p. 22.


Product Source Contact
EndNote Thomson ResearchSoft
Ibid Nota Bene
SigmaPlot 10.0 Systat Software International
Voxler Golden Software