1
 
        INFO & DEC SCI 480                                       SPRING, 1994
        CLUSTER ANALYSIS                                               SCLOVE
 
            OBTAINING "NORMIX", WOLFE'S MIXTURE-MODEL CLUSTERING PROGRAM
             Notes on Chapter 3 (A Review of Clustering Methods), cont'd
 
 
             Wolfe (1971) gives a procedure for testing H0: number of
        clusters is r vs H1: number of clusters is r' > r.
        Let
                       likelihood ratio  =  max L(r)/max L(r'),
 
        where  max L(k)  is the likelihood for k clusters, maximized over the
        unknown  parameters  (i.e.,  with max likelihood estimates put in for
        unknown parameter values).  In many  statistical  problems,  minus  2
        times  the  (natural)  log of the likelihood ratio has a distribution
        which is approximated by a chi-square distribution,  the  d.f.  being
        equal  to the difference in the numbers of parameters.  In this case,
        this distribution is approximated by a constant  times  a  chi-square
        distribution  with  an  appropriate number of d.f.; the constants are
        given in Wolfe (1971).
 
        Obtaining the NORMIX Program
        ---------------------------------------
             The NORMIX program has been converted to run  on  386/486  MSDOS
        systems.   It  uses  extended memory as needed to run large problems.
        The computational procedures are  identical  to  the  1978  mainframe
        version,  but  a  number of improvements have been made to the I/O to
        make it more user-friendly.  In addition,  it  runs  about  25  times
        faster   than  it  did  on  old  mainframes.   A  shareware  version,
        NORMIX20.ZIP, may be downloaded by anonymous ftp from SIMTEL20 mirror
        sites, such as:
 
             SITE                  DIRECTORY
 
        OAK.Oakland.Edu         /pub/msdos/statstic
 
        archive.orst.edu       /pub/mirrors/simtel20
                                 /msdos/statstic
 
        DIRECTIONS:  FTP to one of the above sites, change to the
        appropriate directory using the cd command.
        Put the FTP program into BINARY or IMAGE mode.  Then
 
                                 get NORMIX20.zip
 
             Copy the NORMIX20.zip file to your C: drive root directory.
        Unzip it with
 
               PKUNZIP -D NORMIX20.ZIP,
          or     UNZIP -D NORMIX20.ZIP.
 
             Then follow the directions in the README and documentation
        files.
 
1
 
        IDS 480, S '94                                                  p.  2
        ---------------------------------------------------------------------
 
 
           References
           ----------
 
        Wolfe, J. H. (1970).  Pattern clustering by multivariate mixture
           analysis.  Multivariate Behavioral Research 5, 329-350.
 
        Wolfe, J. H. (1971).  A Monte Carlo study of the sampling
           distribution of the likelihood ratio for mixtures of multinormal
           distributions.  Naval Personnel and Training Research Laboratory
           Technical Bulletin STB 72-2.  San Diego, California.
 
 
        SLS:ss/NORMIX.480s94
        30-Mar-94