plot.lrpsadmmpath: Plotting for 'lrpsadmmpath' Objects

In benjaminfrot/lrpsadmm: Low-Rank Plus Sparse Estimation via Alternating Direction Method of Multipliers (ADMM)

Description

Plots the sparsity and number of edges of S as a function of the tuning parameter lambda (see documentation of lrpsadmmpath). The rank of L as a function of lambda.

The user can also provide a matrix of 0s and 1s to be used as "ground truth". If the location of non-zero entries of S is known (for example because this is simulated data), then the precision/recall metrics are computed and plotted. See examples below.

Usage

## S3 method for class 'lrpsadmmpath'
plot(x, ground.truth = NULL)

Arguments

x	An object of class lrpsadmmpath output by the function lrpsadmmpath
ground.truth	A binary matrix representing the adjacency matrix of the "true" graph one seeks to recover. Useful mostly on simulated data where the true parameter is known.

See Also

lrpsadmm lrpsadmm.cv

Examples

set.seed(0)
# Generate data with a well-powered dataset
sim.data <- generate.latent.ggm.data(n=2000, p=100, h=5, outlier.fraction = 0.0,
                                     sparsity = 0.02, sparsity.latent = 0.7)
X <- sim.data$obs.data; Sigma <- cor(X) # Sample correlation matrix


gamma <- 0.1 # Some reasonble value for gamma
# We ask for 30 lambdas, but the sparse graph becomes too dense so the
# computation is stopped.
my.path <- lrpsadmm.path(Sigma = Sigma, gamma = gamma,
                         lambda.ratio = 1e-03, n.lambdas = 30, verbose = TRUE)

# This time let us ask for 30 values, but let us narrow down the range by using a
# a smaller ratio
my.path <- lrpsadmm.path(Sigma = Sigma, gamma = gamma,
                         lambda.max = 0.96, lambda.ratio = 0.1, n.lambdas = 30, verbose = TRUE)

# Plot some basic information about the path
plot(my.path)
# Look at the first graph in the path
plot(my.path[[1]]$fit)
# Because this is simulated data, we know the ground truth
# Let us use it to compute the precsion and recall metrics
# along the path
ground.truth <- sim.data$precision.matrix[1:100, 1:100]
# Remove the elements along the diagonal. Keep a matrix of 0s and 1s
ground.truth <- 1 * (( ground.truth - diag(diag(ground.truth)) ) !=0)
# There is a new plot with the precision / recall curve
plot(my.path, ground.truth = ground.truth)

benjaminfrot/lrpsadmm documentation built on Oct. 19, 2019, 8:13 a.m.