 |
 |
 |
 |
 |
 |
|
 |
|
|
|
|
|
|
prudsys DISCOVERER - Technology
Architecture
DISCOVERER was written completely in C++ and is a Windows desktop application. The following figure illustrates this architecture:
Both in terms of its implementation and its mathematical methods, DISCOVERER is suited to the analysis of large volumes of data with millions of datasets.
Data import
Data import in DISCOVERER takes place by importing text files using the text import module or by scanning in databases using the database wizard.
Access to databases takes place in the database wizard either as direct access, where the number of possible databases is constantly growing, or via ODBC for all common databases such as Oracle, Sybase, MS SQL-Server or Interbase.
Tables of a relational database are graphically linked in the database wizard.
Analytical functions
The classification module is the heart of DISCOVERER. This is where classifiers in the form of non-linear decision trees are created, visualised and evaluated. This module is the world's most powerful classification package.
The following classification methods are supported:
|
■
|
Axis-parallel decision trees
|
|
■
|
Linear classification methods
|
|
■
|
Non-linear classification methods
|
|
■
|
Sparse grids.
|
All of the classification methods listed above are included in the framework of non-linear decision trees. The visualisation of non-linear decision trees is intuitive and clearly shows their function.
The world's leading optimisation package, XPRESS MP, was also integrated into the multivariate classification method.
The sparse grid methods were developed in conjunction with the University of Bonn under the direction of Prof. Griebel.
DISCOVERER also features a powerful statistics package.
Results and export
The DISCOVERER report module can summarize all graphics and results tables in one report.
In addition to saving classifiers in an internal DISCOVERER format, DISCOVERER also offers other ways of exporting classifiers:
|
■
|
Classifiers can either be exported as SQL-conforming rules to text files.
|
|
■
|
Or, classifiers can also be exported in an extended (non-linear model descriptions) PMML format. The PMML files can be integrated via the  XELOPES library for online classification in third party applications.
|
|
|
|
|
|
|
 |
|
|
|
