Miner3D, release 7
24 January 2008
Reviewed by Felix Grant
Scientific computing for data comprehension is now available to almost everyone in an industrialised society. Blanket access to technology creates a pressure for the benefits of its increasingly widespread application but not the instant specialist training, nor the time, to make best use of it. Symbolic or numerical tools and results, in particular, require either formal expertise or crude black box approaches based on preset boundary conditions. Even where the expertise exists, applying such methods to surveying new data is not the most efficient use of human or computing resources.
Visualisation and sonification, however, are rapidly accepted by the human brain’s own internal “modelling software” as just one more set of perceptual modalities. Miner3D inhabits a relatively new market sector of products explicitly designed from the ground up to exploit this capacity, rapidly tapping into intuitive internalisation of data patterns at a sophisticated level which bypasses formal interpretation. There are four versions available – basic, professional, enterprise (supplied for review and discussed here) and developer – with progressively incremented features.
Data can be taken directly from most common databases (including Access), anything reachable through ODBC, Excel, CSV and tab structured text files, or the Windows clipboard, as well as the native M3D file format. Automatic refresh can be set, reloading from the same source at intervals of between five seconds and five minutes, so that on screen displays reflect changing data. Once imported they can be viewed in a native data sheet, spreadsheet style, or inspected variable by variable in a properties panel, but are by default kept out of sight. A panel of basic statistical descriptors for central tendency and dispersion can also be toggled on and off, as can display of filter criteria and formulae.
Although audiovisual data models can be built manually, allowing Miner3D to automatically provide initial visual assignments is often more useful. The data series showing highest variability are mapped to axes and other major aspects such as colour and marker size as appropriate to the selected plot type, but these cycle through successive combinations with each click of the “build” button. Extremely rapid exploration of the relationships within a very complex data set can be conducted in this way, stopping to manually tune or develop those which look promising. Audio (rendering variable values in terms of frequency, amplitude and stereophonic spatial placement; speech synthesis can also supplement text output) has to be added manually.
Plot types available are scatter and bar (both in either two or three dimensions), line graph, or surface, arranged as straight Cartesian layouts or as neural Kohonen maps, and can be displayed simultaneously in multiple panes stacked or tiled. Trellis displays are also available. Principal component analysis and K-means clustering can be applied with a couple of mouse clicks, and are immediately reflected across the current displays.
Throughout the package, control is though either selectors or (for variable parameters) sliders. A lot of thought has gone into how these are grouped, with dendritic arrangement of individual controls within logical panels under a tabbed ribbon, on a multiple dashboard metaphor. Formatting of graph spaces (planes, axes, boxing, and so on) is probably the best implementation I’ve seen; rotation is likewise the smoothest and most responsive, though a little idiosyncratic.
While its marketing emphasises business use, Miner3D is equally suitable for scientific or engineering use; I briefly tested it with both expert and lay users on their own data sets in ecological, medical, military analysis, nuclear, paper manufacture, pharmaceutical and petrochemical contexts, with equally impressive success.