HCLUSTER <input>,<output_line> / F. Åberg 11.5 1996
Performs hierarchical clustering of observations in the specified data
or on a distance matrix.
HCLUSTER let you plot a dendrogram on CRT or a Postscript printer.
When data is used, the variable with the label should be activated with
letter L, and the variables to compute the distances from with letter A.
If no L activated variable is found, the first activated variable is
used if it is of string type.
If no suitable label is found, the observation numbers are used as labels.
Note: the label variable must be of string (S) type.
The HCLUSTER module recognizes a distance matrix as input when the name
ends with the .MAT extension.
(eg. DIST and DISTV modules by S.Mustonen are useful for making
distance matrices.)
On next screen about various specifications.
Specification Diffrent values Abbrevation Remarks
METHOD SINGLE_LINKAGE SIN or 1 (default)
COMPLETE_LINKAGE COM or 2
AVERAGE_LINKAGE AVE or 3
WEIGHTED_AVERAGE WAV or 4
CENTROID CEN or 5
WEIGHTED_CENTROID WCE OR 6
MINIMUM_VARIANCE MIN or 7 Also called Wards method.
SAVEDIST <matrix> Default: no saving.
<textfile> With extension .TXT
DISTANCE SQUARED_EUCLIDIAN SQU or SQR or 1 (default)
EUCLIDIAN EUC or 2
CITY_BLOCK CIT OR 3
CANBERRA_METRIC CAN or 4
TREEDATA <datafile> Default: #TREE#
Used also for PS file.
RESULTS 0..10 Short output.
>10 Long output.
PLOT PS or POSTSCRIPT Output for PostScript.
PS,LANDSCAPE Print format: Landscape
more specifications on next screen.
SCALING YES ( any value will do. ) Performs standardization
of variables before com-
puting distances.
zero mean, unit variance
WEIGHTS <weight matrix> Vector with weights.
Survo matrix; 1 column
m rows, in the same
order as the activated
variables.
examples on next screen.
HCLUSTER DECA,CUR+1 / METHOD=MINIMUM_VARIANCE SAVEDIST=MAT1
The distance matrix is saved in matrix file MAT1.MAT. If n>90 then distances
are saved as a text file MAT1.TXT
HCLUSTER D.MAT,CUR+1 / TREEDATA=C:TMPTREE1 RESULTS=0
Performs cluster analysis based on distance matrix D. The data that contains
the dendrogram is saved in data file TREE1.SVO in current datapath.
Only the lines relevant for plotting the dendrogram are as output.
Note that TREEDATA and SAVEDIST can include a path name.
HCLUSTER MYDATA,CUR+1 / DISTANCE=CIT PLOT=PS SAVEDIST=DIST1.TXT
Uses method single linkage and the distances are CITY BLOCK measures.
The dendrogram is 'printed' to a PostScript file. The name (and path)
is the same as in TREEDATA but with the .PS extension.
The distance matrix is saved as a textfile in DIST1.TXT (in datapath).
Note that the distance matrix is not saved by default.
More about HCLUSTER on next screen.
The HCLUSTER module uses an agglomerative algorithm.
Other distance measures can be used by making a
distance matrix with the DIST module.
Note that HCLUSTER only work with dissimilarity measures.
Literature used for programming the HCLUSTER module:
Anderberg Michael R. : Cluster Analysis for Applications, NY & London, 1973
Jain Anil K. : Algorithms for Clustering Data, 1988
Everitt Brian S. : Cluster Analysis, 1983
1 = More information on additional multivariate operations
M = More information on multivariate analysis
Help System (web edition)