Log-diagonal representation of a variance matrix
Translate a square symmetric matrix with positive diagonal elements into a vector of the logarithms of the diagonal elements with the correlations as an attribute, and vice versa.
logVarCor(x, corr, ...)
x |
If a matrix, translate into a vector with a "corr" attribute. If a vector, translate into a matrix. |
corr |
optional vector of correlations for the
Use a "corr" attribute of |
... |
(not currently used) |
if(length(dim(x))==2) return log(diag(x))
with an attribute "corr" equal to the
lower.tri of cov2cor(x).
Otherwise, return a covariance matrix from
x as described above.
Spencer Graves
##
## 1. Trivial 1 x 1 matrix
##
# 1.1. convert vector to "matrix"
mat1 <- logVarCor(1)
# check
all.equal(mat1, matrix(exp(1), 1))
# 1.2. Convert 1 x 1 matrix to vector
lVCd1 <- logVarCor(diag(1))
# check
lVCd1. <- 0
attr(lVCd1., 'corr') <- numeric(0)
all.equal(lVCd1, lVCd1.)
##
## 2. simple 2 x 2 matrix
##
# 2.1. convert 1:2 into a matrix
lVC2 <- logVarCor(1:2)
# check
lVC2. <- diag(exp(1:2))
all.equal(lVC2, lVC2.)
# 2.2. Convert a matrix into a vector
lVC2d <- logVarCor(diag(1:2))
# check
lVC2d. <- log(1:2)
attr(lVC2d., 'corr') <- 0
all.equal(lVC2d, lVC2d.)
##
## 3. 3-d covariance matrix with nonzero correlations
##
# 3.1. Create matrix
(ex3 <- tcrossprod(matrix(c(rep(1,3), 0:2), 3)))
dimnames(ex3) <- list(letters[1:3], letters[1:3])
# 3.2. Convert to vector
(Ex3 <- logVarCor(ex3))
# check
Ex3. <- log(c(1, 2, 5))
names(Ex3.) <- letters[1:3]
attr(Ex3., 'corr') <- c(1/sqrt(2), 1/sqrt(5), 3/sqrt(10))
all.equal(Ex3, Ex3.)
# 3.3. Convert back to a matrix
Ex3.2 <- logVarCor(Ex3)
# check
all.equal(ex3, Ex3.2)Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.