07-21-2014 08:00 AM
I am working with colleagues who are creating a series of predictive models using SAS Enterprise Miner and then I am trying to create the same process within SAS Enterprise guide however I am getting different results to my colleagues.
As part of the testing between Miner and EG we have created a very simple model where we are trying to predict a value by only using one input variable (award) from 2012 data. We are not doing any imputing on missing values as award is fully populated. Then the scores from the 2012 data are being applied to 2013 data. Miner is however splitting the data 50/50 into training and validation data to carry out the model work.
I have tried using proc logistic to first create the model scores (using the outmodel option). I then use proc logistic again to calculate the probabilities (using the inmodel statement) however I only get 14% of the results that have a difference of 0.001 or less.
Looking at the code used in Enterprise Miner it uses a procdure called proc dmreg, which does not exist in SAS EG
I have tried using proc reg but because the input variables are character that does not work. I have tried to use proc GLM and GLM select to no avail.
Before using proc logistic again I have tried using the proc transreg procedure to change the character variable into numeric however I still get the same results from my original proc logistic test.
I have tried splitting the data 50/50 before putting it into the proc logistic model but the results were even worse.
Can anyone recommend a SAS procedure that will mimic the SAS Enterprise miner process.
07-21-2014 08:24 AM
If you want to prove correctness yes/no of the Eminer suite than you will get a hard time. It is not easy rebuilding Eminer steps into Eguide.
All intermediate data with Eminer is found in maps on the OS level. You could use those ones. The training/validation data could be copied .
Eminer is doing generating a lot of SAS code using a lot of dedicated metadata-tables (own structure). You could use and run that all be hand but .... why?.
With EGuide code editor the proc dmreg (see miner 4.3 docs) could be used. The requirement is that Eguide is using the same sas-session as Eminer allowed to run licensed miner procs.
07-21-2014 08:29 AM
Miner and SAS EG are running on seperate servers so I cannot use the dmreg procedure.
Thanks for the idea of looking at the the intermediate data found in the OS Level. I'll have to review those so see if that can give me any insight.
07-21-2014 10:34 AM
That is a little bit strange implementation. The eminer license is coming in with both of them expecting some cooperation between those.
Eminer processes actually are running.needing a workspace-server. That is the same as Eguide is using. Modeling result of Eminer can go to the metadata for other components. It is giving my a feeling some conceptual communications in architecture, support, functionality and usage has been bypassed.