Building models with SAS Enterprise Miner, SAS Factory Miner, SAS Visual Data Mining and Machine Learning or just with programming

3 questions regarding to SAS EM association/data partition nodes

Occasional Contributor
Posts: 6

3 questions regarding to SAS EM association/data partition nodes

Hi all, I have following questions regarding to SAS EM and failed to find the answers online... Thanks for your help. 


1). Control point node is to help to simplify the diagram. However, I dont quite understand which result the model node is returning when multiple data sets are linked to it.. For example, if I link 2 data set to 1 model node, result based on which data set will be returned? Or the model node will combine the 2 data sets first and then compute the result?


2). When using Association node for sequence analysis, there are 2 paramters called consolidate time and maximum transaction duration. May I check what's the unit of the 2 fields? For example, if the sequence in my data set is in days and I want to consolidate transactions within 2 days, shall I put "consolidate time = 2 (days) " or "consolidate time = 48 (hours)" ? 


3). Does SAS EM support modification of transacion date set? Or data preparation is supposed to be done using SAS or SAS EG? I am asking as I tried to do data filtering using Filter node, and to do data partition using Data Partition node before doing association rule analysis. However, the filtering or data partition didnt work when I put data type = transaction; What's more, I am not supposed to change data type = raw/train since Association node only works for transactional data. 


I am not sure whether this is the proper forum for such questions. Any suggestion is appreciated! 

Posts: 272

Re: 3 questions regarding to SAS EM association/data partition nodes

For question 1), you can attach multiple Input Data nodes to a modeling node (or to a Control Point node then modeling node as you mention), but you can only have at most one data set of each role connecting to the modeling node.  So you can have a data set with the Role property set to Raw or Train in the Input Data node, one with role Validate, and one with role Test all feeding into a modeling node, and the modeling node will use the data sets according to their role, score each one, and export the scored versions to subsequent nodes, keeping the same role.  Hope that makes sense and helps.

Ask a Question
Discussion stats
  • 1 reply
  • 2 in conversation