Skip to contents

Confusion metrics at a PIP threshold

confusion_at_threshold.Rd

Computes true/false positive/negative counts and rates by comparing a continuous score matrix (e.g., PIP) against a binary ground truth adjacency matrix at a given threshold. Only the upper triangle is used.

Usage

confusion_at_threshold(score_mat, truth_mat, thr, upper_only = TRUE)

Arguments

score_mat

Numeric matrix [p, p]; continuous edge scores (e.g., posterior inclusion probabilities from pip_edges).

truth_mat

Numeric matrix [p, p]; binary ground truth adjacency (1 = edge present, 0 = absent).

thr

Numeric; threshold for edge selection. An edge is selected if score_mat[i,j] >= thr.

upper_only

Logical; if TRUE (default), only the upper triangle is evaluated (appropriate for undirected graphs).

Value

A named numeric vector with components:

TP

True positives (correctly selected edges).

FP

False positives (incorrectly selected non-edges).

TN

True negatives (correctly excluded non-edges).

FN

False negatives (missed true edges).

TPR

True positive rate (sensitivity / recall).

FPR

False positive rate (1 - specificity).

See also

roc_auc(), pip_edges()

Examples

sim <- simulate_multiggm(K = 2, p = 8, n = 80, seed = 1)
fit <- multiggm_mcmc(data_list = sim$data_list, burnin = 200, nsave = 100)
pip <- pip_edges(fit)
confusion_at_threshold(pip[,,1], sim$adj_list[[1]], thr = 0.5)
#>         TP         FP         TN         FN        TPR        FPR 
#>  8.0000000  0.0000000 15.0000000  5.0000000  0.6153846  0.0000000