Probabilistic opponent-model search in Bao

In Probabilistic Opponent-Model search (PrOM search) the opponent is modelled by a mixed strategy of N opponent types omega(0) ... omega(N-1). The opponent is assumed to adopt at every move one of the opponent types 0, according to the probability Pr(omega(i)). We hypothesize that PrOM search is a better search mechanism than Opponent-Model search (OM search) and Minimax search. In this paper we investigate two questions: (1) to which extent is PrOM search better than OM search and Minimax search in the game of Bao? and (2) which opponent type is most advantageous to use? To answer the second question we constructed Five evaluation functions which we applied in a tournament consisting of 352,000 games. Our conclusions are twofold: (1) in Bao, PrOM search performs better than OM search and sometimes also better than Minimax search even when no perfect information of the opponent is available, and (2) for an adequate performance of PrOM search, emphasis on the own evaluation function in the opponent model should be higher than assumed so far.