Compute comprehensive evaluation metrics for an ssjgl fit

Given an ssjgl fit object and the true adjacency/precision matrices, computes edge-level metrics (TPR, FPR, AUC) and estimation error (Frobenius norm, KL divergence).

Usage

compute_metrics(
  fit,
  true_adj,
  true_omega = NULL,
  v0_index = NULL,
  threshold = 0.5
)

Arguments

fit

An object of class ssjgl.

true_adj

List of K binary adjacency matrices (true graph structure).

true_omega

List of K true precision matrices. Default NULL.

v0_index

Integer index into the v0 ladder. Default NULL uses the last step.

threshold

Threshold for binarizing edge probabilities. Default 0.5.

Value

A list with elements:

per_group

List of K sublists, each with TP, FP, TN, FN, TPR, FPR, precision, F1, frobenius_norm, and KL_divergence.

roc

List of K ROC objects (from roc_auc).

overall

Named list with mean TPR, mean FPR, mean AUC across groups.

Details

The edge indicator in SSJGL is shared by all groups, so every group is scored with the same PIP matrix (against its own true graph).

Examples

sim <- simulate_ssjgl_data(K = 2, p = 10, n = 50, seed = 1)
fit <- ssjgl(sim$data_list, penalty = "fused",
             lambda0 = 1, lambda1 = 0.5, lambda2 = 0.5,
             v0s = 0.01, maxitr.em = 10, impute = FALSE)
m <- compute_metrics(fit, sim$adj_list, sim$Omega_list)
m$overall
#> $mean_TPR
#> [1] 0.9428105
#> 
#> $mean_FPR
#> [1] 0.3452381
#> 
#> $mean_AUC
#> [1] 0.8783112
#>