6. Betting Strategy

Using the game state, opponent models and hand evaluation derived from that data, the betting strategy determines whether to fold, call/check, or bet/raise. But exactly what information is useful and how should it be used? The answer to these questions are not trivial and this is one of the reasons that poker is a good testbed for artificial intelligence. One approach would be to use the game theoretic optimal betting strategy, but, despite the fact that it is very complex to calculate, human opponents do not play optimally so this may not be the best decision in practice (leading to the most profit).

A minimal system could simply be based on hand strength ( i.e. ignore hand potential and simply bet/call based on the immediate strength of our hand). Refinements to the betting strategy would involve the addition of high-level strategy concepts (like slowplaying or bluffing). For each decision to be made, one of several variables (like PPOT) is compared to some threshold (which may be based on another variable, like $pot\_odds$). This structure uses expert knowledge but is easy to implement and the overhead is insignificant. All of these refinements are intended to be quick ways to select the play which (hopefully) has the largest expected value (EV), since computing the exact EV is not feasible. There is a real-time constraint, in that a single game of poker only takes a few minutes to play, and from Figure 4.2 we see that the game tree can be very large.

This chapter first describes the pre-flop betting strategy (which is treated as a special case). This is followed by an explanation of a simple post-flop (flop, turn and river) betting strategy (using only hand strength) which serves as a template, and then the measures and strategies that Loki uses for the post-flop rounds. Finally, unimplemented strategies are discussed.

Note that this chapter is included for completeness. The betting strategy is (so far, necessarily) ad hoc and therefore left undeveloped (it is a target for future improvement with computer-oriented techniques such as Monte Carlo simulation).