Representing poker decisions as a set of three probabilities provides a suitable infrastructure to perform well in a noisy environment, where randomized strategies and misinformation are important aspects of strong play. Besides the performance improvement obtained by the use of PTs as a non-deterministic betting strategy and as the reweighting factor in the opponent modeling module, the PT generation function improves the design of the system by encapsulating all the knowledge-based components of Loki-2 in a single routine.
Although the results obtained are encouraging, there are still
opportunities for improvement. Using showdown and bluffing information about
the opponents to perform more accurate specific opponent modeling,
replacing Loki-1's preflop betting strategy with a PT-based strategy,
using PTs in preflop reweighting,
and refining the knowledge in the current probability triple function
are some of the possible next steps.