Offline Data Mining

Hi,

I am currently working for a client that is interested in performing some data mining on their customer data. The plan was to build a data warehouse, cubes, RS reports etc. and serve this up over the web.

However, we now have the problem of having to cater for countries with a very low bandwidth where the Reporting Services approach would not deliver the required performance due to the dependency of the solution on the network bandwith.

I am therefore looking into processing offline cubes and copying them over the network as part of the overnight process to a server in the countries.

I noticed that PivotTable Services in 2000 had the CREATE MINING MODEL ...... DDL statement for creating offline mining models.

Is this capability still there with SSAS 2005?

Can anyone post any links to usefull resources on this subject?

Thanks in advance.

There a couple of ways to deal with offline scenarios:

1. AS 2005 does support creation of local mining models if you are connected to a .cub file (in the "local" mode formerly known as PivotTable Services) through the MSOLAP.3 OLE DB provider. Only the Decision Trees and Clustering algorithms are supported in this case. You can then export these models individually, ship the files over to your server and import them back into your server. See http://www.sqlserverdatamining.com/DMCommunity/TipsNTricks/129.aspx for more details.

2. If the clients have AS 2005 Standard Edition, they can create models using all algorithms and export as described above. In addition, you can also export server models, ship them over to the client and import them back in, allowing local querying without going over the network.

|||

Yes, the capability is still there in SQL Server 2005.

This page contains some information about client scenarios with Analysis Services 2005, including local mining models: http://msdn2.microsoft.com/en-us/library/ms174518.aspx

More details specific for working with local mining models: http://msdn2.microsoft.com/en-us/library/ms345148.aspx#sqldmprgrm_topic7

The local mining models are limited in functionality. Major limitations:

- only 2 algorithms are supported (decision trees and clustering)

- BI Dev Studio cannot be used to explore local mining models

- server models cannot be exported and then imported in local cubes

|||Good stuff. Thanks for the help guys.sql