DATA Step, Macro, Functions and more

Scoring Codes after Proc Cluster rules found

Accepted Solution Solved
Reply
Contributor
Posts: 46
Accepted Solution

Scoring Codes after Proc Cluster rules found

Hello All, 

I'd like to learn if there is a scoring statement for Proc Custer?

so after I determine the Proc Cluster segment rules, I'd like to apply the "rules" (restuls from outtree=?) onto new dataset (same people with diffrent X generated with seperate time) to see if the cluster memmebrship stays the same.

 

Is there any "scoring codes" for doing this work?

 

thank you for your help in advance!

 

 

 


Accepted Solutions
Solution
‎04-25-2016 10:06 AM
Respected Advisor
Posts: 4,651

Re: Scoring Codes after Proc Cluster rules found

Look at proc fastclus with options outstat= and instat=. First step, do the clustering on original dataset and output cluster statistics with outstat=. Second step, call proc fastclus again with another dataset and use option instat= to bring back the cluster definitions produced with the original dataset. The second step does assignment only.

PG

View solution in original post


All Replies
Solution
‎04-25-2016 10:06 AM
Respected Advisor
Posts: 4,651

Re: Scoring Codes after Proc Cluster rules found

Look at proc fastclus with options outstat= and instat=. First step, do the clustering on original dataset and output cluster statistics with outstat=. Second step, call proc fastclus again with another dataset and use option instat= to bring back the cluster definitions produced with the original dataset. The second step does assignment only.

PG
Contributor
Posts: 46

Re: Scoring Codes after Proc Cluster rules found

 

I'd like to know the scoring procedure specifically for Proc Cluster. thank you!

Super User
Posts: 9,681

Re: Scoring Codes after Proc Cluster rules found

That would lead you to DISCRIM Analysis ,Not CLUSTER Analysis, Check PROC DISCRIM .
Contributor
Posts: 46

Re: Scoring Codes after Proc Cluster rules found

so is there no way to score after "model building" by proc cluster?

Super User
Posts: 9,681

Re: Scoring Codes after Proc Cluster rules found

Yes. That is DISCRIM Analysis thing. How do you know your test data would be properly clustered ? The difference thing between DISCRIM Analysis and Cluster Analysis is DISCRIM has a TRAIN dataset, CLUSTER Analysis don't have.
☑ This topic is SOLVED.

Need further help from the community? Please ask a new question.

Discussion stats
  • 5 replies
  • 353 views
  • 1 like
  • 3 in conversation