Caret package findCorrelation() function

Question

Hello I am having trouble with the findCorrelation() function, Here is my input and the output:

findCorrelation(train, cutoff = .50, verbose = FALSE)

Error in findCorrelation_exact(x = x, cutoff = cutoff, verbose = verbose) : correlation matrix is not symmetric

Does anyone know why this is happening?

Solution 1

the findCorrelation function expects a correlation matrix as x value, so try this instead:

findCorrelation(cor(train), cutoff = .50, verbose = FALSE)

Reference: Caret pre-processing

Solution 2

Well, it happens because the matrix possibly does not have as many columns as rows (or vice versa). E.g.

library(caret)
train <- cor(mtcars)
findCorrelation(train, cutoff = .50, verbose = FALSE)
# works
findCorrelation(train[, -1], cutoff = .50, verbose = FALSE)
# Error in findCorrelation_exact(x = x, cutoff = cutoff, verbose = verbose) : 
#   correlation matrix is not symmetric
dim(train[, -1])
# [1] 11 10

(At least that would be my guess according to the error message.)

Solution 3

If you are using a data frame then you may have to make it into a matrix first.

corval <- findCorrelation(cor(as.matrix(train)), cutoff = 0.80, verbose = FALSE, names = FALSE)

Comments

[wotter]

I tend to agree 😀