To make a betting decision Loki-1 uses an evaluation function to determine which of the three actions (fold, check/call, or bet/raise) is more likely to be profitable. In fact, the ``best'' action can be considered as the action with the highest expected value over all the possible scenarios. However, Loki-1 provides too much information to the opponents if it always takes the best action. Loki-1 is predictable and the opponents exploit that predictability. We use probability triples, a set of three probabilities representing the three types of actions, to provide a randomized betting strategy for Loki-2 and to represent the probabilistic nature of poker. This representation isolates the expert knowledge in a single function (probability triple generation routine) and allows a computer oriented (i.e. easy to maintain and modify) design of the evaluation function.
A probability triple (PT) is an ordered list of three values, PT = (f,c,r) such that f + c + r = 1.0. Each value represents the likelihood that the next action in a given state is a fold (f), a call (c), or a raise (r), respectively. Probability triples are used in three places in Loki-2: 1) as a stand-alone betting strategy, 2) as the reweighting factor in the opponent modeling module, and 3) as the action generation mechanism during the simulations.