This presentation is an extension of my review paper on stochastic block models (SBMs). There is this ever-present issue of selecting the number of blocks (or groups) when applying an SBM. This presentation focuses on the approaches that use a criterion to select this number. Various criteria that existed then are:
- Likelihood modularity
- Complete data log-likelihood
- Integrated complete data log-likelihood (ICL)
- Approximate ICL
- Observed data log-likelihood
- Approximate observed data log-likelihood
- Marginal log-likelihood
- Bayesian information criterion