Abstract:
For
Markov random fields with finite set of states, a modification of the
Bayesian information criterion admits strongly consistent estimation of
the smallest region that determines the conditional distributions. Phase
transition or nonstationarity do not affect the result.