Model Selection
The number of templates is selected using WAIC, held-out link prediction, and consensus stability. Held-out topology prediction evaluates posterior mean edge-presence probabilities by ROC AUC. Stability assessment aligns templates across i…
1 sources - 4 claims
The number of templates is selected using WAIC, held-out link prediction, and consensus stability. Held-out topology prediction evaluates posterior mean edge-presence probabilities by ROC AUC. Stability assessment aligns templates across independent runs to handle label switching. WAIC should not be used alone because it may mildly decrease as template count increases.