GeminiBPath {jointMeanCov}

Estimate Row-Row Covariance Using Gemini for a Sequence of Penalties

Description

GeminiBPath estimates the row-row covariance, inverse covariance, correlation, and inverse correlation matrices using Gemini with a sequence of penalty parameters. For identifiability, the covariance factors A and B are scaled so that A has trace m, where m is the number of columns of X, A is the column-column covariance matrix, and B is the row-row covariance matrix.

Usage

GeminiBPath(X, rowpen.list, penalize.diagonal = FALSE)

Arguments

X

Data matrix, of dimensions n by m.

rowpen.list

Vector of penalty parameters, should be increasing (analogous to the glassopath function of the glasso package).

penalize.diagonal

Logical indicating whether to penalize the off-diagonal entries of the correlation matrix. Default is FALSE.

Value

corr.B.hat

array of estimated correlation matrices, of dimension (nrow(X), nrow(X), length(rowpen.list)).

corr.B.hat.inv

array of estimated inverse correlation matrices, of dimension (nrow(X), nrow(X), length(rowpen.list)).

B.hat

array of estimated covariance matrices, of dimension (nrow(X), nrow(X), length(rowpen.list)).

B.hat.inv

array of estimated inverse covariance matrices, of dimension (nrow(X), nrow(X), length(rowpen.list)).

Examples

# Generate a data matrix.
n1 <- 5
n2 <- 5
n <- n1 + n2
m <- 20
X <- matrix(rnorm(n * m), nrow=n, ncol=m)

# Apply GeminiBPath for a sequence of penalty parameters.
rowpen.list <- sqrt(log(m) / n) * c(1, 0.5, 0.1)
out <- GeminiBPath(X, rowpen.list, penalize.diagonal=FALSE)

# Display the estimated correlation matrix corresponding
# to penalty 0.1, rounded to two decimal places.
print(round(out$corr.B.hat[, , 3], 2))