KMEANS_CLUSTER: Partitioning Observations Into Clusters Based on the Nearest Mean Value

How to:

The KMEANS_CLUSTER function partitions observations into a specified number of clusters based on the nearest mean value. The function returns the cluster number assigned to the field value passed as a parameter.

Note: If there are not enough points to create the number of clusters requested, the value -10 is returned for any cluster that cannot be created.

Syntax: How to Partition Observations Into Clusters Based on the Nearest Mean Value

KMEANS_CLUSTER(number, percent, iterations, tolerance, 
        [prefix1.]field1[, [prefix1.]field2 ...])

where:

number

Integer

Is number of clusters to extract.

percent

Numeric

Is the percent of training set size (the percent of the total data to use in the calculations). The default value is AUTO, which uses the internal default percent.

iterations

Integer

Is the maximum number of times to recalculate using the means previously generated. The default value is AUTO, which uses the internal default number of iterations.

tolerance

Numeric

Is a weight value between zero (0) and 1.0. The value AUTO uses the internal default tolerance.

prefix1, prefix2

Defines an optional aggregation operator to apply to the field before using it in the calculation. Valid operators are:

  • SUM. which calculates the sum of the field values. SUM is the default value.
  • CNT. which calculates a count of the field values.
  • AVE. which calculates the average of the field values.
  • MIN. which calculates the minimum of the field values.
  • MAX. which calculates the maximum of the field values.
  • FST. which retrieves the first value of the field.
  • LST. which retrieves the last value of the field.

Note: The operators PCT., RPCT., TOT., MDN., MDE., RNK., and DST. are not supported.

field1

Numeric

Is the set of data to be analyzed.

field2

Numeric

Is an optional set of data to be analyzed.

Example: Partitioning Data Values Into Clusters

The following request partitions the DOLLARS field values into four clusters and displays the result as a scatter chart in which the color represents the cluster. The request uses the default values for the percent, iterations, and tolerance parameters by passing them as the value 0 (zero).

SET PARTITION_ON = PENULTIMATE
GRAPH FILE GGSALES
PRINT UNITS DOLLARS
COMPUTE KMEAN1/D20.2 TITLE 'K-MEANS'=  KMEANS_CLUSTER(4, AUTO, AUTO, AUTO, DOLLARS);
ON GRAPH SET LOOKGRAPH SCATTER
ON GRAPH PCHOLD FORMAT JSCHART
ON GRAPH SET STYLE *
INCLUDE=IBFS:/FILE/IBI_HTML_DIR/ibi_themes/Warm.sty,$
type = data, column = N2, bucket=y-axis,$
type=data, column= N1, bucket=x-axis,$
type=data, column=N3, bucket=color,$
GRID=OFF,$
*GRAPH_JS_FINAL
colorScale: {
		colorMode: 'discrete',
		colorBands: [{start: 1, stop: 1.99, color: 'red'}, {start: 2, stop: 2.99, color: 'green'}, 
               {start: 3, stop: 3.99, color: 'yellow'}, {start: 3.99, stop: 4, color: 'blue'} ]
	}
*END
ENDSTYLE
END

The output is shown in the following image.