Parallax Software for Multidimensional Visualization

by Alfred Inselberg, aiisreal at tau.ac.il
the inventor of Parallel Coordinates

Here is a multi-dimensional visualization
Click to see a larger picture

The software is based on Parallel Coordinates which is a methodology for the unambiguous (i.e. no loss of information) visualization of multivariate data and RELATIONS. The discovery of multivariate/multidimensional relations in a dataset is transformed into a 2-D pattern recognition problem. The software's unique interface, queries, and boolean operators enable the visual/interactive discovery of complex relations in multivariate datasets, and in turn finding the effect these relations have on various objectives. Unexpected relations have been discovered in datasets with more than 100 variables from which sensitivities, repetitive patterns, other trends and salient properties are found. The visualization not only helps the discovery process but ALSO the presentation and EXPLANATION of the results.

Recently very efficient classifiers based on the methodology were found and have now been implemented in the software. Specifically, let a dataset consist of N categories (i.e. subsets). Here the classifiers :

1. Discover explicit rules (if they exist -- i.e. if there is sufficient information in the data) which distinguish a category from the others.
2. Find the MINIMAL subset of parameters which suffices to specify the rule WITHOUT LOSS OF INFORMATION (i.e. this is NOT an approximation). -- this has achieved tremendous i.e. 1 to 5 or more reduction in the dimensionality of the problem
3. Order these variables so as to optimize the separation between the categories -- this provides a very useful rating of the importance of the parameters.
4. Provides the rule VISUALLY.
5. Provides information on the geometrical distribution of the data.

Price, ordering and technical information

Please contact Alfred Inselberg, aiisreal at tau.ac.il, and mention that you saw it in KDnuggets.