Optional Syntax Elements for TD_KMeans - Teradata Vantage

The following are the hyperparameters that you tune when using the TD_KMeans function:

ModelTable

Specifies the ModelTable name to save the clustering data model. If specified, then a model containing centroids of clusters is saved in the specified ModelTable name.

InitialCentroidsMethod

Specifies the initialization method to be used for selecting initial set of centroids. Not required, if the InitialCentroidsTable is specified. Allowed values: [random, kmeans++].

'random': The initial set of centroids are selected randomly.

'kmeans++': The initial set of centroids are selected using the KMeans++ algorithm.

Default: 'random'

NumClusters

Specifies the number of clusters to create from the clustering data. Not required, if the InitialCentroidsTable is specified.

Seed

Specifies a non-negative integer value to randomly select the initial cluster centroid positions from the input table rows. Not required, if the InitialCentroidsTable is specified.

StopThreshold

The algorithm converges if the distance between the centroids from the previous iteration and the current iteration is less than the specified value.

Default: 0.0395

MaxIterNum

Specifies the maximum number of iterations for the K-means algorithm. The algorithm stops after performing the specified number of iterations even if the convergence criterion is not met.

Default: 10

NumInit

Specifies the number of times to repeat clustering with different initial centroid seeds. The function returns the model having the least value of Total Within Cluster Squared Sum.

Not required, if the InitialCentroidsTable is specified.

Default: 1

OutputClusterAssignment

Specifies whether to return the Cluster Assignment information.

Default: False