Sunday, March 11, 2012

Datamining a Olap Cube

I'm new to datamining and have gather some knowledge by following the tutorial that comes with sql server 2005.
I have following created a association mining model and tested it by using various DMX queries. I works quite satisfactory.

But I would like to use this mining model on a cube. Ie a calculated measure that would show top 1 recommendation for my customer portfolio.
My knowlede reaches to the extent that you need to feed data by using the OpenQuery or OpenRowset queries.
I would like to "stick" my model to the cube on some way and the extract values in the same manner as filtering criterias in stead of manually generating DMX queries.

I have googled a bit and searched MSDN without luck. Maybe I'm looking at this all wrong.

Is this possible?

Regards Bergur

Could this help?

http://forums.microsoft.com/MSDN/ShowPost.aspx?PostID=1640504&SiteID=1

It presents an example of how to create a custom action containing a data mining prediction. Not really a calculated measure, but still a way to associate a prediction result with an cube cell.

Alternately, you could try this more complex approach:

- have your transaction data in the same cube

- build an OLAP mining model based on the cube

- use the MDXPredict MDX function to retrieve predictions from a mining model inside the calculated measure