Posted by arlene

New Mining-Model Algorithms

Analysis Services 2005 features seven data-mining algorithms—five more than in the 2000 version. In addition, the original two algorithms, decision trees and clustering, have been updated.

The new algorithms are listed below:

1. Association rules—Used to create a set of rules used in predictions. Most useful in making predictions against large amounts of transactional sales data.
2. Time series—Used to predict trends; can be useful when working with financial data, such as stock prices.
3. Naive Bayes—Used only against noncontinuous variables (for example, a product name) and therefore performs very quickly.
4. Sequence clustering—In addition to grouping similar data into clusters, it uses sequence analysis to determine the order in which events occur.

5. Neural nets—Based on an AI technique, this is useful for determining things like whether a customer is good or bad. It is the most thorough algorithm and therefore the most time consuming.

Mining Data in Real Time

Analysis Services 2005 will continue to support the method of processing data used in the 2000 version. This method is a “pull” method in which data used to process the model is pulled from the data source at the time it is processed. For most situations this is all right, since data mining is generally used to extract meaning from historical data that does not change all that much. Also, mining involves looking for trends in the data and not querying for specific values.

With the new version, you can now use a push method to retrieve data from a Data Transformation Services (DTS) package or a custom application. Another option, in between the two, is to use a proactive cache when you are working with data from an OLAP data source. In this scenario, data is refreshed based on predefined parameters, such as the amount of time between data pulls.

Migrating a Mining Model Created with SQL Server 2000

A migration wizard is included with Analysis Services 2005 that allows you to migrate a mining database created with Analysis Services 2000. You still have to preprocess the mining models once they are migrated, but at least you do not have to recreate them from scratch. The limitations on migrating cubes as of the beta 2 version include not being able to migrate remote partitions and linked cubes. Linked cubes have been replaced with linked measure groups.

Individual mining models can also be copied to Analysis Services 2005 by using a PMML (Predictive Model Markup Language) query. You can then create a mining model in Analysis Services 2005 by using the Create Mining Model statement and referencing the PMML retrieved. This method does not copy the bindings, though, so you will only be able to view the content in Analysis Services 2005 and not be able to reprocess the model. This could be useful, however, if you want a quick way to view the results of an old mining model using the new tools in Analysis Services 2005.

Longhorn

Longhorn is the code name for the next big Microsoft operating system. The final client version of this operating system is not expected to arrive before 2006, but when it does arrive, it should offer computing advances that can be utilized by many intelligent-based applications (the server version is expected to be released sometime in 2007).

The client version will include two pillars known as Avalon and Indigo. Avalon is the presentation layer, and Indigo is the communications layer. The initial release should be followed by the release of a new storage system known as WinFS as a beta version. WinFS represents a significant change in the way data is organized and accessed by applications.

Possibly related posts: (automatically generated)
Future Programming, the Next Development Platform part 2

• Network and Programming: the .Net framework
• Future Programming, the Next Development Platform part 3
• Keeping Multiple Database Applications in Sync with Replication
• Future Programming, the Next Development Platform part 1
• Key Differences Between Unix/Linux and NetWare (Applications)
• The Active Directory Schema
• Website Hosting Sever, some Pitfalls you need to avoid part 1
• Speech Application Programming Interface Speech-Related Technologies
• Scripting Database Objects
• Network and Servers Technical Compare continue...
• The Skype API Operates on Windows and Mae OS X