Model Manager


Model Manager

We are exploring the use of Model Manager to manage our model deployment. The models will be developed in a partner tool and the PMML imported into Model Manager. I understand that you have to link the deployed model to an existing analytic data set that is defined in Warehouse Miner. Since we're only using the data set in Warehouse Miner to score the model, rather than develop it, can we only define in that set of data elements required for scoring?

Our analytic data set used by the partner tool is very large and we're trying to avoid re-defining it in Warehouse Miner, if possible.



Re: Model Manager

Yes, you will just need to create in Teradata an analytical data set (ADS) consisting of the variables that the model is dependent upon. The ADS can be created by hand coding SQL to generate the table that will be populated with all the required variables for the model at run time. Another approach would be to use Teradata ADS Generator to create the ADS in Teradata. The PMML from the partner tool is consumed (imported) by Teradata Warehouse Miner and converted to SQL. From Model Manager the user will select the ADS and model for model deployment..
Keep in mind that Model Manager is included in the Teradata Warehouse Miner software that that Nationwide has licensed, but in order to use Model Manager Nationwide will need to license Teradata Application Platform (TAP).

Bob Juhasz
Teradata Warehouse Miner
Product Manager

Re: Model Manager

I have a layman question, what is the difference between Teradata Warehouse Miner and ADS Generator?

Also does Model Manager is packaged with Teradata Warehouse Miner? or its a separate component?

Re: Model Manager

The question is asking about the difference between Teradata Warehouse Miner and Teradata Analytical Data Set (ADS).

For brevity's sake, the ADS may be (and often is the case) a subset or data-cube taken from the whole. An ADS may contain data of customers which have contacted Customer Service for electronic issues in the last six months. It may be used and retained since the data may be required multiple times. This would cut down the constant building of the data-cube. i've heard it referenced as a behavior-based entity. The more specific it can be defined the better.

The Warehouse Miner allows for queries without data movement. At least, that's one of the features I appreciate. It allows query tuning and retries that could result in what goes into an ADS.