caret (version 7.0-1)

findLinearCombos: Determine linear combinations in a matrix

Description

Enumerate and resolve the linear combinations in a numeric matrix

Usage

findLinearCombos(x)

Value

a list with elements:

linearCombos

If there are linear combinations, this will be a list with elements for each dependency that contains vectors of column numbers.

remove

a list of column numbers that can be removed to counter the linear combinations

Arguments

x

a numeric matrix

Author

Kirk Mettler and Jed Wing (enumLC) and Max Kuhn (findLinearCombos)

Details

The QR decomposition is used to determine if the matrix is full rank and then identify the sets of columns that are involved in the dependencies.

To "resolve" them, columns are iteratively removed and the matrix rank is rechecked.

Examples

testData1 <- matrix(0, nrow=20, ncol=8)
testData1[,1] <- 1
testData1[,2] <- round(rnorm(20), 1)
testData1[,3] <- round(rnorm(20), 1)
testData1[,4] <- round(rnorm(20), 1)
testData1[,5] <- 0.5 * testData1[,2] - 0.25 * testData1[,3] - 0.25 * testData1[,4]
testData1[1:4,6] <- 1
testData1[5:10,7] <- 1
testData1[11:20,8] <- 1

findLinearCombos(testData1)

testData2 <- matrix(0, nrow=6, ncol=6)
testData2[,1] <- c(1, 1, 1, 1, 1, 1)
testData2[,2] <- c(1, 1, 1, 0, 0, 0)
testData2[,3] <- c(0, 0, 0, 1, 1, 1)
testData2[,4] <- c(1, 0, 0, 1, 0, 0)
testData2[,5] <- c(0, 1, 0, 0, 1, 0)
testData2[,6] <- c(0, 0, 1, 0, 0, 1)

findLinearCombos(testData2)