Numero (version 1.2.0)

nroPrune: Reduce collinearity within a dataset

Description

Detect and merge collinear columns into first principal components.

Usage

nroPrune(data, modules)

Arguments

data

A matrix or a data frame.

modules

Pruning parameter, see details.

Value

A data frame or a matrix where a module of collinear columns has been replaced by a single column. The aggregated values are linear combinations of the module columns; the coefficients define the principal component of the module data.

The output also contains the attribute "modules", which can be passed to the function to replicate the same pruning procedure for another dataset.

Details

The pruning parameter modules is an integer that sets the desired number of columns in the pruned dataset. If necessary, the number is automatically revised if the original value cannot be applied to the dataset.

The input argument modules can also be a list object that is attached to the output of a previous call to the function, see the description of the return value.

To determine modules of collinear variables, the function uses K-means clustering with Spearman correlation as the distance metric.

Examples

# NOT RUN {
# Import data.
fname <- system.file("extdata", "finndiane.txt", package = "Numero")
dataset <- read.delim(file = fname)

# Split into men and women.
ds.men <- dataset[which(dataset$MALE == 1),]
ds.women <- dataset[which(dataset$MALE == 0),]

# Exclude unusable columns.
ds.men$INDEX <- NULL
ds.women$INDEX <- NULL
ds.men$MALE <- NULL
ds.women$MALE <- NULL

# Merge collinear variables in one dataset according to the other.
results.men <- nroPrune(data = ds.men, modules = 3)
results.women <- nroPrune(data = ds.women, modules = results.men)
print(attr(results.men, "modules"))
print(summary(results.men$MODULE.1))
print(summary(results.women$MODULE.1))
# }