I have created a Propensity Model using Logistic Regression in Enterprise Miner 5.2.
The process has created a model based on the Training data set and when I use Score Node in the workflow I receive the Score Code. My question is, Is there a way we can implement the scoring in database programming language. ex: I have SQL SERVER 2005 database in which I have the full set of data on which I would like to apply the REGRESSION MODEL created in SAS enterprise Miner, How can I do that?
I don't believe that you could just convert Miner code to SQL Server SQL, not with a reasonable effort anyway. But I'm pretty sure that you could apply your model to SQL Server data through a SAS/ACCESS libname.
As seen on the web, SAS is creating modules that would make possible to run scoring models inside external RDBMS. Teradata is the first implementation (SAS Scoring Accelerator for Teradata). Haven't heard anything about a SQL Server implementation.
I think that you can apply it in SQL/Server, though I've not tried it.
The scoring just applies the logistic regression formula to the new data. The basic output of a binary logistic regression is
alpha + beta*x
where alpha is the intercept and the beta's are the regression coefficients.
If you want the risk probability for a new vector x', then the formula is
exp(alpha + beta*x')/(1 + exp(alpha + beta*x') )
where "exp(alpha + beta*x')" is the natural number "e" raised to the "alpha + beta*x'" power.
Thanks a lot.
After designing the workflow in ENTERPRISE MINER for logistic regression, I ran the workflow and got the score code.
Is alpha + beta*x value available in Score code script which is generated?
Is exp(alpha + beta*x')/(1 + exp(alpha + beta*x') ) the score value for each record?
I am new to SAS Enterprise Miner can you please help me out?
I don't use enterprise miner, so I can't answer the specific questions. I know you get the outputs in PROC LOGISTIC in SAS/Stat. The score is either the alpha + beta*x (also known as the logit) or the predicted probability.
You may need to get a text and bone up on logistic regression or work with a statistician before you get too deep into this. At least read the chapter on logistic regression in the SAS/Stat manual.