variable_cluster: Cluster a set of numeric variables

In dnegrey/miscTools: R package with miscellaneous tools

Description

variable_cluster() performs non-hierarchical (disjoint) variable clustering on the numeric variables in a specified data frame. This approach is similar to that of the varclus procedure in SAS. Like that approach, the default behavior is to remove records with any missing values (na.rm = TRUE). Unlike that approach, however, missing values can be easily replaced with their column means (na.rm = FALSE). Note that this can greatly increase the size of the data to be clustered, which will increase the run time of the function. An example of this was timed on a data frame with about 230K observations and 200 variables where the function took 10 minutes. Future efforts should be made to optimize this or explore alternative approaches.

Usage

variable_cluster(x, n, na.rm = TRUE)

Arguments

x	data frame; contains the variables to be clustered
n	integer value >= 2; number of desired clusters
na.rm	logical value; should records with missing values be removed?

Value

A data frame with class "mt_variable_cluster" containing the following columns:

• VarName: variable name (character)

• PrimaryCluster: assigned cluster (integer)

• RsquaredToPrimaryCluster: R-squared to primary cluster (numeric)

• NearestCluster: next nearest cluster (integer)

• RsquaredToNearestCluster: R-squared to nearest cluster (numeric)

• OneMinusRsquaredRatio: one minus R-squared ratio (numeric)

See Also

kmeansvar

Examples

variable_cluster(mtcars, 3)

dnegrey/miscTools documentation built on May 3, 2019, 2:57 p.m.