Data Mining slicing / filtering

Is there a way to dynamically slice / filter data when running a data mining query?

(Dynamically means - not deployed as part of the structure)

For example - I’d like find associations to products sold in different geographic locations.

Does the answer differ if the structure is based on a cube or on a rational DB?

The answer is no - the counts of product associations are the counts for the products with which the model was changed. You would have to create specific structure for each subset of the data you would want to model.

You may be able to use many-many dimensions in OLAP and then provide the appropriate slice to see association counts by region, but you won't have a mining model that can perform predictions, and you will only see pairwise counts.

Data Mining slicing / filtering

Is there a way to dynamically slice / filter data when running a data mining query?

(Dynamically means - not deployed as part of the structure)

For example - I’d like find associations to products sold in different geographic locations.

Does the answer differ if the structure is based on a cube or on a rational DB?

The answer is no - the counts of product associations are the counts for the products with which the model was changed. You would have to create specific structure for each subset of the data you would want to model.

You may be able to use many-many dimensions in OLAP and then provide the appropriate slice to see association counts by region, but you won't have a mining model that can perform predictions, and you will only see pairwise counts.

Data Mining in analyzing servers

I am trying to setup a means to analyze our servers using SQL Server 2005 Data Mining. I will be collecting all the logs from the Event Log of each server and store them in a SQL Server database for analysis.

Has anybody done this before? Can you give some pointers? I would appreciate your assistance.

I've thought about this problem a bit, but not had the time to implement any solution. Generally my thoughts are around defining a "failure" variable, and then using log information to predict that failure, or to predict time to failure.

"Failure" in this case could be anything, for example

Users can't connect