I trained a Liar Dice game (similar rule please refer to https://www.wikihow.com/Play-Liar%27s-Dice , and i am using 2 player + 1 dice vs 1 dice) with FSICFR algorithm and got an information set with relative data (eg. sigma, sumSigma, regret...). But I don't know how to use this data in the dice game?
So far, i try to use the action with maximum probability(of the sigma) as the next move of the game when i provided the current "game state -- player's dice rank" and "action(claim) history", but the result is not good enough (it tend to "CHALLENGE" the opponent's CLAIM).
my question is:
what is better way to use this data of the trained information set? my goal is try to guess the opponent's dice rank according his claim history and get the best next move.
how can i update the data of the trained information set in real-time when the algorithm play the game?
could anyone give me some advice or any idea?
thanks in advance!