Providing software solutions since 1976

• Log in
• Create profile
• Welcome
  [Log out]
• Edit profile

• Print  |
• Feedback  |
• Share

  • E-mail

  
  
  

  E-mail This Page

  Your Name:*
  Your E-mail:*
  Recipient E-mail:*
  Send a copy to me
  Personal Message:
  
  

  * indicates a required field

There are two ways. One uses PROC FASTCLUS, the other uses PROC DISCRIM.

PROC FASTCLUS:

First, compute the means from each of your current clusters to use as a SEED= data set in PROC FASTCLUS. Then run FASTCLUS with DATA= your new data set, SEED= the data set of means from your current clusters, MAXITER=0, REPLACE=NONE, and MAXC= the number of clusters you have. Be sure to use a VAR statement listing the same variables as you used in PROC CLUSTER.

The FASTCLUS method is suitable if the original clusters were obtained by FASTCLUS or by CLUSTER with METHOD=WARD. With other methods, there is no formal justification for using FASTCLUS, although results will probably be reasonable with METHOD=AVERAGE or CENTROID, and maybe with MEDIAN and FLEXIBLE. For METHOD=COMPLETE, you would want to use FASTCLUS with LEAST=MAX.

PROC DISCRIM:

Use your data set of cluster-identified observations as the DATA= set and your data set of new observations as the TESTDATA= data set. List the cluster-identifying variable in the CLASS statement. Use the same VAR statement as you used in PROC CLUSTER. See the example titled Linear Discriminant Analysis of Remote-Sensing Data on Crops in the DISCRIM chapter of the SAS/STAT User's Guide.

DISCRIM with METHOD=NPAR is suitable if the original clusters were obtained by MODECLUS or by CLUSTER with METHOD= DENSITY or TWOSTAGE. Be sure to specify the same density-estimation method in DISCRIM as was used for clustering. DISCRIM using nearest neighbor (K=1) would be the natural choice for clusters from METHOD=SINGLE. DISCRIM with METHOD=NORMAL is problematic because the distributional assumptions in DISCRIM do not correspond to any of the clustering methods.

NOTE ON REPLICATING A FASTCLUS CLUSTERING:

With the default settings, FASTCLUS will offer a quick solution for even very large data sets. The default convergence value of 0.02 allows a reasonable solution in a reasonable amount of time. Changing the convergence criterion to 0 may require a large number of iterations and substantially increase computation time.

By default, FASTCLUS gives no information about whether or not a run converged as it presumes you want the fast clustering. However, if you specify a value of MAXITER= that is greater than one, the procedure presumes that you are interested in reaching convergence so the SAS log will show

WARNING: Iteration limit reached without convergence  

RELEASES PRIOR TO VERSION 7:

If you want to create clusters using FASTCLUS that you can exactly reproduce later, then on the initial FASTCLUS run, you will need to specify CONVERGE=0 and increase MAXITER= until it converges. Then, running FASTCLUS as at the beginning of this paper article will reproduce the same clusters.

BEGINNING WITH VERSION 7

If you want to create clusters using FASTCLUS that you can exactly reproduce later, then on the initial FASTCLUS run, create an OUTSTAT= data set. (You do not have to specify CONVERGE=0.) Then to create the exact same clusters on the second FASTCLUS run, specify the OUTSTAT data set from the first run as the INSTAT= data set. You do not have to specify MAXC=, REPLACE= or MAXITER= the second time if you use this method.

Operating System and Release Information