Algorithm used

This is where past research ends. The philosophy behind this section is to show that a search strategy can be inferred. The search strategy of a chess master is a lot more complicated than the ones used here, and hence further study could find closer matches than the methods in this thesis. What is planned, is to show that by enumerating strategies, a closest match can be found. The process used here is rather simple. It involves enumerating a selection of strategies for being able to deduce which of them is used to generate the game inferred. There were five strategies chosen here but there is no reason for the number of strategies to be restricted. The main reason these five where chosen is to demonstrate that even with only a small number of strategies, it is possible to differentiate between them using this method. The strategies are restricted to methods that are not purely optimisation techniques they actually have an impact upon the evaluation of the best move for a particular opponent. This means strategies such as alpha beta search, transposition tables, and other methods that are used to speed up the minimax algorithm cannot be used. These strategies do not involve the use of extra information, but rather they remove redundant information. These techniques are used for speeding up the search and don't affect the answer, thus, by purely looking at the data, there is no way of telling the difference between them.

The strategies chosen are one ply, two ply, one ply with a maximum of two ply quiescent, one ply with a maximum of three ply quiescent, and one ply with a maximum of four ply quiescent. All these will generate different results. The same probabilistic approach is again used so that the inference is more realistic, and the moves played in the inference do not match exactly that of the original. This avoiding of exact matches is more realistic as people overall do not always play in the same manner.

The method of inference is as follows: The evaluation function is kept fixed and only the strategy is changed. The chess program will first generate different numbers of games ranging from 1, 2, 5 and 10 games. The program, then stores this information in lists and makes an inference of these games after the correct number of passes. A given strategy is first chosen and then a game is played using that technique (Source), the moves in the Source are then used for each of the strategies (Evaluators). Each of the Evaluators assesses the moves of the Source and places this information in their game tree (a list). The inference program then finds the message length of each of the five different Evaluators and infers the most likely strategy to have been used by the Source, as the Evaluator with the lowest message length. The information for the message lengths and inferred technique is then saved to a file. Each strategy has its own message length and as a result it makes it possible to differentiate between them and, if need be, give the probability of a chosen strategy being true.