Deploying Miner Models in SQL Server Environment (PMML, SQL, Model Manager?)

[waynefield]

Hi,

So, we are beginning to develop a suite of models in Miner, based on input variables coming from our SQL Server warehouse.

I want to explore our options for scalable deployment, which I believe are:

1. Ad hoc export of scores -> import into relevant systems (eg. CRM platform)
2. Export PMML code -> how can SQL Server understand this
3. Use Model Manager

At the moment we use option 1, which is clunky, time consuming and not particularly scalable.

In the short term I'd like to explore option 2.   Is PMML the right way to go?   Can SQL Server parse this, or do we need to convert directly to SQL (and is this even possible from within EM)?

In the medium term I'd like to configure a Model Manager environment.   However I am told that SQL Server cannot support in database model scoring, so I am not sure how to get around this?

Any support or guidance on the above would be vastly appreciated.

Thanks,

Wayne

[Reeza]

SQL server supports R/Enterprise R which should support PMMl fully to score your data. 

Depending on your final model you may or may not be able to convert it to SQL. Decision trees can be easily converted. 

[waynefield]

Thanks - so can I export directly from EM into PMML format (without needing to use R)?

[jsmithers]

Hi Wayne,

The plan would be to export to PMML from Model Manager and then import to SQL server either on to Analysis services or to R services.

Thanks