Scale Example 5: Use Scale Output in KMeans - Teradata Vantage

This example uses the Scale function to scale data (using the maxabs method) before inputting it to the function KMeans, which outputs the centroids of the clusters in the data set. (KMeans explains the reason for scaling data before inputting it to a distance-based analysis function like KMeans.)

For an example of using Scale functions to convert input variables into z-scores for use with the Principal Components Analysis (PCA) functions, see PCA Example.

Input

• input: computers_train1, as in KMeans Example 1: NumClusters, UnpackColumns ('false') by Default
• statistic: created by calling the ScaleMap function inside the Scale call

SQL Call to Create Table of Scaled Data

CREATE MULTISET TABLE computers_normalized AS (
  SELECT * FROM Scale (
    ON computers_train1 AS "input" PARTITION BY ANY
    ON (
      SELECT * FROM ScaleMap (
        ON computers_train1
        USING
        TargetColumns ('[1:5]')
        MissValue ('omit')
      ) AS dt1
    ) AS statistic DIMENSION
    USING
    ScaleMethod ('maxabs')
    Accumulate ('id')
  ) AS dt2
) WITH DATA;

SQL Call to Input Scaled Data to KMeans

SELECT * FROM KMeans (
  ON computers_normalized AS InputTable
  OUT TABLE OutputTable (computers_centroid)
  USING
  NumClusters ('8')
  StopThreshold ('0.05')
  MaxIterNum ('10')
) AS dt;

Output