Batch Processing of SAS Miner Diagram

[arindam1984]

Hi,

I am trying to schedule an entire SAS Miner diagram which has both training and scoring parts. I have extracted the SAS code from the last node. The extracted code runs fine in SAS EG. For linear regression, all the model parameters are regenerated when I run the extracted code (as the model gets retrained). However there is one hard coded variable for 'Decmeta Data Set' and 'Decdata Data Set'. It is for the target variable. On reading, it seems that it has got to do with prior probabilities. However on manually checking the response rate of the training data, the numbers do not match. Moreover if I run the same data with a different model, then the number changes.

My question is, what does this variable do? Since it is hard coded, can I derive it manually and replace it when I schedule it via SAS DI? Any help will be great.

Thanks

Arindam

[DougWielenga]

I have extracted the SAS code from the last node. The extracted code runs fine in SAS EG. For linear regression, all the model parameters are regenerated when I run the extracted code (as the model gets retrained). However there is one hard coded variable for 'Decmeta Data Set' and 'Decdata Data Set'. It is for the target variable. On reading, it seems that it has got to do with prior probabilities. However on manually checking the response rate of the training data, the numbers do not match. Moreover if I run the same data with a different model, then the number changes.

My question is, what does this variable do?

"DEC" is from "Decision" and contains information which is used to assign each observation to a particular outcome (or decision). From the SAS Enterprise Miner help in the SAS Enterprise Miner Macro Variables Reference

Use This Option	To Specify This.	Default Value
columnsmeta=	columnsmeta data set
decdata=	name of the decision data set, if decision matrix is used	WORK.DECDATA
decmeta=	name of the decision metadata data set, if decision matrix is used	WORK.DECMETA
dectype=	type of decision (LOSS or PROFIT), if decision matrix is used

The DECDATA data set contains the Decision outcomes as well as 'weights' that can be specified by the user in a Decision Profile in order to control how an observation is assigned to a prediction class during scoring.   Batch Processing is a method by which the same process can be used to train a different data set using the same approach which includes any decision information that was associated with the diagram.   While the data is hard-coded, it could be altered manually as desired, but it would be far safer to simply alter the decision profile prior to obtaining the batch code.   Please note that it is not necessary to run a flow prior to accessing the batch code, so it should not be necessary to create or alter any of these data sets manually (and I would strongly recommend against doing so). 

One of the advantages of SAS Enterprise Miner is that you define how variables are to be used (e.g. Measurement Level, Role, etc... ) so that processing can continue without having to specify this for each new process.  The information about how a variable is to be used is stored as metadata which you can change in the middle of a flow using a Metadata node.  The DECMETA data merely contains metadata information about the variables in the DECDATA data set. 

Hope this helps!

Cordially,

Doug 

[arindam1984]

Thanks. But my flow does not have a decision node currently (it had once before but not now when I extracted the batch code). I used the batch code to check what would get executed in SAS Model Manager. My concern is, it is using numbers that I do not want. I have raised a SAS Track for this.