rhogrid {EMgaussian}R Documentation

Create sequence of possible tuning parameter values

Description

Create sequence of possible tuning parameter values

Usage

rhogrid(
  n.rho,
  method = c("qgraph", "glassopath"),
  rho.min.ratio = 0.01,
  dat = NULL,
  S = NULL,
  ...
)

Arguments

n.rho

Integer corresponding to the number of tuning parameter values.

method

Character corresponding to the method to create tuning parameter values ("qgraph" or "glassopath"); see Details.

rho.min.ratio

Numeric value that mimics EBICglasso behavior for tuning parameter. i.e., "Ratio of lowest (tuning parameter) compared to maximal (tuning parameter)".

dat

The raw data, if S is not provided used to estimate S. Not required if S is provided.

S

Covariance matrix for the data. If provided, supersedes dat.

...

Other arguments passed down to em.prec.

Details

For regularized estimation of the Gaussian graphical model, a sequence or grid of possible tuning parameter values is often tried, with the tuning parameter that optimizes some criterion (EBIC, k-fold cross validation) chosen. This is an attempt to automate some of the sequence creation. Code is borrowed from qgraph and glasso, acknowledged in references below. Both require some estimate of the covariance matrix in order to do regularization; if not provided, em.prec with default options is attempted.

For "qgraph" the max value is determined by the maximum absolute value of the difference between the covariance matrix and an identity matrix. The min is rho.min.ratio times the max value. A sequence that is equally spaced on a log scale is then constructed between these two values.

For "glassopath", the max value is the max absolute value of the covariance matrix. The min is the max divided by the number of desired tuning parameter values. A sequence that is equally spaced between these two values is then constructed.

Value

A vector of possible tuning parameter values.

References

qgraph: Epskamp, S., Cramer, A., Waldorp, L., Schmittmann, V. D., & Borsboom, D. (2012). qgraph: Network visualizations of relationships in psychometric data. Journal of Statistical Software, 48 (1), 1-18.

glasso (i.e., glassopath option): Friedman, J. H., Hastie, T., & Tibshirani, R. (2014). glasso: Graphical lasso estimation of Gaussian graphical models. Retrieved from https://CRAN.R-project.org/package=glasso

Examples


library(psych)
data(bfi)

# pick 50 values using the approach qgraph uses; give data as input 
rho <- rhogrid(50, method="qgraph", dat = bfi[,1:25])

emresult <- em.cov(bfi[,1:25])
S<-emresult$S

# pick 50 values using the approach glasso uses; give S as input
rho2 <- rhogrid(50, method="glassopath", S = S)



[Package EMgaussian version 0.2.1 Index]