Interfacing to other software packages

Data format conversion

It is important that the user can use the library with data acquired by other sources. To transform data sets to SPRANNLIB readable format a program dataimp can be used. The user must make minor changes to the source code, which determines the format of the data to be transformed. The data format of SPRANNLIB data sets is straight forward and transformation from SPRLIB format to a generic form is easily done by Unix grep and tail commands.

Interfacing to graphics software

the graphical display of data that is obtained from the training sets or simulation runs of the library, is an important way to obtain good insight in the nature of the data. Since there are numerous excellent graphics packages available, it was chosen to interface to these packages, rather than writing the graphics ourselves. A tool is available to convert network and training data to formats that can be processed by spreadsheets or business graphics packages and to Mathematica [Wolfram 1992] (see figure 4.1). Another output format is an image that is determined by changing two network inputs and computing the corresponding output. Such image can then be processed by an image processing package.

  
Figure 4.1: Interfaces to Mathematica provide facilities to display data sets and network behaviour. In the left figure a simple data set is displayed; the large and small dots represent two different classes in a two dimensional feature space. In the right figure, the network output as a function of two inputs is sketched.

Interfacing to statistical software

A similar interface as was mentioned above was developed to convert data to formats of statistical software packages. This facilitates a convenient way to obtain various statistical measures of the data. There are several Matlab [Matlab 1992] m-files available with which SPRANNLIB datafiles can be read, there is also an m-file that processess neural network description files (see the appendix for a description of the format).