AS 2005 is an intensely compelling product. If I may borrow a concept from OLAP, the Analysis Services product improves on its predecessor in a number of “dimensions,” including ease of use, feature richness, sophistication of client tools, and programmability. This release truly feels like a serious tool for developers, statisticians, and scientists. It makes AS 2000 look like a bare-bones, proof-of-concept product by comparison.
In preparation for some upcoming presentations (including one at Tech*Ed in a month and a half), I’ve been working especially hard over the last week studying Analysis Services Data Mining and, on the OLAP side, ADO MD.NET programming for both the server and the client. My general impression while examining all of this technology is that it is rich, elegantly extensible, and much easier to use than its predecessors. Some highlights follow:
- ADO MD.NET will be familiar to ADO MD and ADO.NET programmers alike. Do you prefer thinking in terms of axes and cellsets? It’s all covered. More comfortable using DataReaders and DataSets? That’s accommodated too. Jealous of the CLR programmability on the relational side of SQL Server 2005? Don’t be: you can now use the objects in the Microsoft.AnalysisServices.AdomdServer namespace in a .NET assembly to create MDX-callable functions that can accept and return sets, tuples, and scalar values, all with the same security model as on the relational SQLCLR side.
- The number of Data Mining Algorithms has been extended significantly and the ability for you to create your own plug-in algorithms is carefully documented, rather than merely being implemented.
- Tools for browsing models have been extended significantly and are available from both Analysis Services Projects in Visual Studio (a.k.a. Business Intelligence Development Studio) and directly from SQL Server Management Studio. Tools for visual development of prediction queries are also readily available in both tools and are easy to use (in AS 2000, the Prediction Query Wizard is buried in DTS, and viewing prediction query results is rather inconvenient).
- Mining Structures, a new entity in AS 2005 databases, contain a common data source/input column set/predictable column set configuration that can be used by multiple models within the structure. Each constituent model can be built using a different algorithm, the same algorithm with different parameters, or a combination of both. Better yet, new “lift” charts allow you to graph the relative accuracy of all your models, both for predictable columns in general and specific values of those columns. As with other tools, lift charts are available from both Analysis Services projects in Visual Studio and from within SQL Server Management Studio. Here’s a screenshot of the latter (forgive the large size, but I want you to be able to read all the text within):
I continue to believe that the real story to SQL Server 2005 is its business intelligence feature set, and I don’t just mean Reporting Services and ReportBuilder. OLAP and Data Mining technologies from Microsoft are powerful and eminently useful analysis tools. Their potential for adding value to the entire Microsoft stack, from Office to SharePoint, BizTalk to Commerce Server and every member of the Microsoft Business Solutions family, is tremendous. Continue to ignore these technologies at your own peril.