As a result from the list of how an optimal solution should look like, it was decided to build a library of statistical pattern recognition and neural network functions. The final simulation environment consists of a compiled C-library with such functions. Portability is ensured by writing the package in ANSI C code. For its good support of software development and wide use in the scientific community, Unix was chosen as the platform to develop the library.
A standardized set of powerful data structures offers the flexibility that is required
for various types of networks, together with an easy access of its parameters. The large
set of routines operates on these data structures. The user has to write his or her own
main program, which calls the library functions.
One of the drawbacks of the followed approach is that there is no graphical user
interface that is available in other environments. This was partially solved by writing
interfaces to packages for data analysis and graphics facilities. Typical interfaces are
available for Mathematica 
[Wolfram 1992], plot and graph for Unix
systems, Matlab [Matlab 1992], spreadsheets and business graphics software. For
statistical pattern recognition a similar library, called SPRLIB [Schmidt 1992], was
developed and the artificial neural network library (ANNLIB) became a part of it. These
two separately developed libraries have been merged in to the currently available SPRANNLIB
package.