JGL_AIC_surfaceplot: 3D plot of AIC values

In GiulioCostantini/JGL2: JGL2: Joint Graphical Lasso, select the lambda parameters using AIC or crossvalidation

Description

Compute the criterioin defined by aicfun on ncand l1 x ncand l2 pairs of l1 and l2. Optionally does a 3d plot of the criterion on the l1 and l2 values

Usage

JGL_AIC_surfaceplot(dat, splt, l1min = 0, l1max = 1, l2min = 0, l2max = 1, ncand = 20, aicfun = AIC_jgl, plot = TRUE, ncores = 1, ...)

Arguments

dat	A dataset that includes the variables on which the gaussian graphical models should be computed, plus an additional factor "splt" which defines different classes.
splt	Character string. The name of the variable in dat that defines different classes
l1min	Numeric. Minimum value of the lasso parameter lambda1
l1max	Numeric. Maximum value of the lasso parameter lambda1
l2min	Numeric. Minimum value of the lasso parameter lambda2
l2max	Numeric. Maximum value of the lasso parameter lambda2
ncand	Integer. number of values for lambda 1 and for lambda 2
aicfun	AIC_jgl or another function that take as input (1) the output of JGL, (2) n = vector of sample sizes, (3) S = list of covariance matrices and give in output a criterion to minimize
plot	if TRUE make a 3d surface plot of the aicfun values over the l1 and l2 parameters, using function scatter3d
ncores	Number of cores to use. The function is optimized for parallel computing in Windows, parallel computing may not work on other systems.
...	Additional parameters for JGL

Value

A dataframe with the following objects

lambda1	Values of lambda 1 considered
lambda2	Values of lambda 2 considered
aic	The AIC or other criterion for each pair of lambda 1 and lambda 2 values

Author(s)

Giulio Costantini

Examples

## Not run: 
# # a toy example
N <- 1000 # sample size
sigma1 <- matrix(c(1, .5, 0, 0,
                  .5, 1, .2, 0,
                  0, .2, 1, 0,
                  0, 0, 0, 1), ncol = 4)

sigma2 <- matrix(c(1, .5, .4, .4,
                   .5, 1, .2, 0,
                   .4, .2, 1, 0,
                   .4, 0, 0, 1), ncol = 4)


dat <- list()
dat[[1]] <- MASS::mvrnorm(n = N, mu = rep(0, ncol(sigma1)), Sigma = sigma1)
dat[[2]] <- MASS::mvrnorm(n = N, mu = rep(0, ncol(sigma2)), Sigma = sigma2)
lapply(dat, function(x) corpcor::cor2pcor(cor(x)))
dat <- data.frame(rbind(dat[[1]], dat[[2]]))
dat$splt <- c(rep(1, N), rep(2, N))

aics <- JGL_AIC_surfaceplot(dat, splt = "splt", l1max = .01, l2max = .01, ncand = 10, ncores = 7, plot = TRUE)
scatter3d(aic ~ lambda1 + lambda2, data = aics, fit = "smooth", surface = FALSE)

## End(Not run)

GiulioCostantini/JGL2 documentation built on May 6, 2019, 6:29 p.m.