WATERFORD, Conn. A government contractor to the U.S. Navy is reworking its fuzzy-logic development system for use with Microsoft Corp.'s Excel spreadsheet. Sonalysts Inc. said tying its Fuzzy Query system to the widely used spreadsheet will gear its product for a mass audience.
Fuzzy Query for Excel lets users create membership functions that define a semantic relationship between data and the reports generated from it. For instance, if grade point average is a variable and a semantic relationship is established to define a "good" GPA, then a fuzzy membership function could be defined as 100 percent at 4.0 and 0 percent at 3.0, with a smooth sigmoid defining the degree of "goodness" between a GPA of 3.0 and 4.0.
A standard set of predefined fuzzy-set shapes is provided to achieve quick definition of fuzzy sets, including sigmoids, triangles and various step functions. For more complex or idiosyncratic fuzzy membership functions, an Edit Fuzzy Set dialog box permits users to alter the shape of predefined fuzzy sets.
An edit option is also provided by which fuzzy membership functions can be defined from scratch, using a piecewise linear approach that makes it possible to define a multipoint shape. Multipoint shapes let users create uniquely shaped fuzzy membership functions to accurately represent the semantic meaning of concepts.
Once the basic fuzzy membership functions are defined for an application, the user creates a fuzzy query file using the query-controls screen. Here the user defines the relationship among the just-defined fuzzy membership functions.
For instance, if the user defined a fuzzy membership function for "good" GPAs and "good" school attendance, then on the query-controls screen he could define a query that combines the two: A "good" student, overall, could be defined as one who has good grades and good attendance.
The query-controls screens also permits Fuzzy Query for Excel to "weight" the relationship among fuzzy variables. For instance, in the good-student application, if grade point average was to be preferred over attendance, then the "average AND" function could be used to weight GPA as, say, twice as important as attendance.
Once a fuzzy query is thus defined, it can be run on the spreadsheet data, resulting in a ranked list of candidates who meet the fuzzy-queries specifications and, it is to be hoped, accurately represent the semantic relationships defined in the fuzzy membership functions.
This sort of ranked list cannot be achieved without fuzzy logic, according to Sonalysts (Waterford, Conn.), because the "goodness" of each student is defined by his or her relationship to the other students in the data set, rather than by a numerical relationship defined in the query.
Sherlock Holmes, for example, might be considered a marginally good student when compared with the near-perfect performers in this data set. If, however, Holmes were among a group of poorly performing students, then his performance could scoot up to the top of the list. This is achieved without redesigning the application in any way, something that is not possible when using non-fuzzy queries.
Fuzzy Query is also available from Sonalysts as a standalone application. Both versions can be downloaded from the company's Web site.