Combining the evaluations by different patterns

Many patterns can apply to the same state, that is, the moves of more than one pattern have been made. It would be possible to construct complicated rules for combining their evaluations; e.g. the average of the evaluations of applying patterns weighted with a function of the pattern size. One should note that play out is the most time intensive part of the algorithm and combining happens at every play-out move so it has to be fast. The proposed combining rule is:

• The maximum of evaluations given by each pattern that applies but is not a subpattern of another applying pattern.

The motivation is that the more specific pattern gives more accurate information about a move than its subpattern. In Figure 4, patterns 1, 2, and 4 apply to the example state, but pattern 1 is a subpattern of the other two so it is not taken into account. Using the maximum helps in computational complexity. One can take the maximum first within a pattern and then over different patterns. The evaluations of a pattern can be stored in a heap to make the time complexity logarithmic with respect to the number of available moves.