SAS modeling and scoring automatisation

While developing our new FinWEB Customer Online Targeting product, I had a lot of interactions with SAS support (especially SAS Switzerland). Although helpful, most answers were only relevant when using SAS in what I call “standard” mode. For example, when I asked how I could add a macro variable as input parameter to some node in Enterprise Miner (EM), the answer was “why don’t you generate code from EM and then modify the SAS code to add your macro variable?”.

At that point I understood that the way we use SAS is not straightforward. We have worked a lot to automate the use of SAS and I think some readers of Data Mining Research may be interested to see, not the details, but the big picture. I distinguish mainly four levels of automatisation for modeling and scoring in SAS.

1. Use of SAS in standard mode

This is the most standard use of SAS. The data miner creates a model and scores users from the EM GUI.


2. Use of SAS in scoring batch mode

This use is common when the data miner wants to build the model himself but wants to generate the scores automatically.


3. Use of SAS in modeling batch mode

This mode allows to dynamically generate both models and scores. The data miner must still run these procedures manually.


4. Use of SAS in scheduled mode

This is the final automatisation level we have achieved with FinWEB. The data miner builds the model workflow (diagram) and imports it in the system. According to client order, FinWEB automatically schedules model creation and scoring. No code has to be run manually.


This is how we use SAS at FinScore, but this is not the only way. I would love to hear how you use SAS in your company, whether it is automated or not.


Recommended Reading