Compute graph recovery metrics (single group)

Compares an estimated precision matrix to the truth, computing edge-level metrics (TP, FP, TN, FN), sum of squared errors, L1 norm, and Kullback-Leibler divergence.

Usage

getmetric(est, truth, graph)

Arguments

est

Estimated precision matrix (p x p).

truth

True precision matrix (p x p).

graph

Binary adjacency matrix of the true graph (p x p, 0 diagonal).

Value

A named list with elements:

SSE

Sum of squared errors between est and truth (off-diagonal).

L1

L1 norm of the estimated precision (off-diagonal).

nedges

Number of non-zero edges in the estimate.

tp

True positives (edges correctly identified).

fp

False positives (spurious edges).

fn

False negatives (missed edges).

tn

True negatives (correctly absent edges).

dKL

Kullback-Leibler divergence KL(est || truth).

Details

This is a legacy function from the reference implementation. For new code, use compute_metrics which handles multi-group fits directly.

See also

compute_metrics for the recommended multi-group interface.

Examples

# Single-group metric computation
p <- 5
truth <- diag(p)
truth[1,2] <- truth[2,1] <- -0.3
est <- truth + matrix(rnorm(p*p, 0, 0.01), p, p)
est <- (est + t(est)) / 2
graph <- (truth != 0) * 1L; diag(graph) <- 0L
getmetric(est, truth, graph)
#> $SSE
#> [1] 0.0008671386
#> 
#> $L1
#> [1] 0.6981802
#> 
#> $nedges
#> [1] 10
#> 
#> $tp
#> [1] 1
#> 
#> $fp
#> [1] 9
#> 
#> $fn
#> [1] 0
#> 
#> $tn
#> [1] 0
#> 
#> $dKL
#> [1] 2.500406
#>