The prudsys XELOPES (eXtEnded Library fOr Prudsys Embedded Solutions) is a platform and data source independent business intelligence library which unites classical data mining methods and new real time analytics. The library can be used as standalone software, offering pre-fabricated solutions to fundamental analytics problems; furthermore, it can be integrated into other software products, emphasising its full performance capacity as an embedded analytical tool. Especially when it comes to new and complex problems, the numerous algorithms of the prudsys XELOPES, which can be combined in modules, allow for the development of adequate solutions.
Data mining standards
prudsys XELOPES supports essential BI standards. This includes CWM, an OMG standard, which specifies the fundamental class structure for a business warehouse in UML, and PMML, an XML format to serialise and exchange data mining models. Both standards enable interoperability between different data mining providers in particular. The current XELOPES implementation is available in several programming languages. The primary implementation is in Java, offering the best possibilities to be integrated into various web and server applications. The PMML standard is supported by all classical data mining models and was expanded by prudsys for the new real time analytics process.
Since classical data mining processes must generally handle extremely large data matrices, the streaming concept for data access was implemented in the prudsys XELOPES. This means that the required data is not necessarily loaded into the main memory but can be read and processed in real time. The data can be in a wide variety of formats. Access to the data is consistently modelled by way of an abstract class. Derived from that, there are completed implementations to process Excel, CSV and log files, including zipped formats, databases, multidimensional data providers etc. The series of data access classes can be easily expanded. To guarantee random access with smaller data volumes, there are special options to store and activate data in temporary storage.
The prudsys XELOPES combines a number of classical data mining models. This applies to decision trees, neuronal networks, SVMs, cluster methods, shopping cart and sequence analysis algorithms as well as prudsys' own methods such as non-linear decision trees and sparse grids. In addition, there is also a statistics package in prudsys XELOPES, which contains fundamental tools for calculating distributions and confidence intervals, sophisticated mathematical methods for solving systems of equations, packages for effective calculation with sparsely populated matrices and tensors as well as an OLAP package, enabling reporting on multidimensional data. Probably the most pioneering and thus outstanding benefit is the so-called agent framework, an interface for the prudsys XELOPES multi-faceted real time analytics.
Based on access of artificial intelligence, a sophisticated development process was used to design a framework offering universal access to all real time methods. At the centre of this framework is a so-called agent that learns from the stimulation of its environment and can then act accordingly. This approach can be transferred for use with product recommendations, automated pricing, disposition and real time scoring. The agent also unites an adaptive model, updated in real time. The agent learns and adapts its model based on the reward, which is high when it comes to a purchased recommendation. Figure 1 symbolises the learn and apply step. The corresponding agents such as RecommAgent (recommendations by way of reinforcement learning), CollaborativeDemandAgent (automatic calculation of order amounts and times), ClickOrderAgent (automatic price adjustment based on clicking patterns) etc. are already integrated into the prudsys XELOPES where the amount of these agents can easily be expanded by way of a flexible deployment mechanism.