Package {netsimhelpers}

Simulate data from a Gaussian Graphical Model (GGM)

Description

Simulate multivariate normal data from a Gaussian Graphical Model (GGM) specified by a partial correlation matrix. Optionally introduces missing values completely at random (MCAR).

Usage

GGMsim(n_obs, omega, missing_prop = 0)

Arguments

Value

A matrix with n_obs rows and ncol(omega) columns. Values are simulated from a multivariate normal distribution implied by the provided partial correlation matrix omega with optional NAs.

Examples

# Construct a 5 node network
omega <- matrix(0, 5, 5)
omega[1, 2] <- omega[2, 1] <- 0.3
omega[2, 3] <- omega[3, 2] <- 0.4

# Simulate data
data <- GGMsim(n_obs = 100, omega = omega)

# Simulate data with missing values
data_missing <- GGMsim(n_obs = 100, omega = omega, missing_prop = 0.1)

Evaluate two networks against each other

Description

Compute performance metrics for an estimated network relative to a known ground-truth network to assess estimation quality. Calculates how well the estimated network recovers the true network structure (which edges exist) and edge weights (how strong connections are).

Usage

evalNet(true, est, metric = NULL, directed = FALSE)

Arguments

Value

A named list of evaluation metrics. Higher values generally indicate better recovery, except for bias metrics (lower is better). Returns NA for metrics that cannot be computed (e.g., when no edges are present).

See Also

genIsing, genGVAR for generating networks

Examples

# Generate true GVAR networks
true_nets <- genGVAR(n_node = 6)

# Simulate estimated networks with some errors
est_nets <- true_nets

# Add false positive to contemporaneous network
est_nets$PCC[2, 5] <- est_nets$PCC[5, 2] <- 0.15

# Remove true edge
true_edge <- which(true_nets$PCC != 0, arr.ind = TRUE)[1, ]
est_nets$PCC[true_edge[1], true_edge[2]] <- 0
est_nets$PCC[true_edge[2], true_edge[1]] <- 0

# Evaluate contemporaneous network recovery
results_contemp <- evalNet(true = true_nets$PCC, est = est_nets$PCC)

# Inspect specific metrics
results_contemp$sensitivity
results_contemp$specificity

# Evaluate temporal network recovery
results_temp <- evalNet(true = true_nets$PDC, est = est_nets$PDC, directed = TRUE)
results_temp$in_strength_correlation

# Evaluate specific metrics only
evalNet(true_nets$PCC, est_nets$PCC,
        metric = c("sensitivity", "precision", "correlation"))

Generate a synthetic GVAR network

Description

Generate network structures for Graphical Vector Autoregression (GVAR) model. Generates both contemporaneous and temporal networks.

Usage

genGVAR(
  n_node,
  contemp_prop_pos = 0.8,
  temp_weight_mean = 0.3,
  temp_weight_sd = 0.1,
  ...
)

Arguments

Value

A list with four elements:

kappa

Precision matrix (inverse covariance) for contemporaneous network

PCC

Partial correlation matrix for contemporaneous network

beta

Temporal regression coefficients matrix

PDC

Partial directed correlations for temporal network

Examples

# Generate a 5-node GVAR network
gvar_net <- genGVAR(n_node = 5)

# Check the networks
gvar_net$PCC   # contemporaneous network
gvar_net$PDC  # temporal network

# More positive edges in contemporaneous network
gvar_net2 <- genGVAR(n_node = 5, contemp_prop_pos = 0.9)

# Stronger temporal effects
gvar_net3 <- genGVAR(n_node = 5, temp_weight_mean = 0.5, temp_weight_sd = 0.2)

Generate a synthetic Ising network

Description

Generate undirected Ising network structures and assign edge weights from a normal distribution.

Usage

genIsing(
  n_node,
  rewire = 0,
  inclusion = NULL,
  weight_mean = 0.3,
  weight_sd = 0.1,
  ...
)

Arguments

Value

A symmetric numeric matrix of size n_node by n_node with zeros on the diagonal. Off-diagonal entries are Ising interaction parameters (0 indicates absence of an edge)

Examples

# Generate a 5 node Ising model with default settings
ising1 <- genIsing(n_node = 5)

# Using edge inclusion probability
ising2 <- genIsing(n_node = 5, inclusion = 0.3)

# Change edge weight distribution
ising3 <- genIsing(n_node = 5, inclusion = 0.4,
                   weight_mean = 0.5, weight_sd = 0.2)