Chessmaster inference

Games in both Fischer and Kasparov's career were chosen and were then employed using the above algorithm to find the search depth. The assumption is, that the search depth can be found, if the estimation of the evaluation function is close to the proper values but not necessarily exactly. This is based on the ability of the previous trial, which found that it is possible to find the evaluation function and the search depth given artificial data. The ability of the algorithm to find, if not the correct depth, the next closest, is also a valuable property. The games chosen were for Kasparov: Kasparov-Pert R. G., 1997, Oakham (England) and Pert N.-Kasparov, 1997 Oakham (England) both at Simultan; and for Fischer were: Fischer-Soltis, 1971 and Kevitz-Fischer 1971, both were at the Manhattan blitz. The main reason for the small number of games is that even after four, the memory needed extends well into the gigabyte range, and even with virtual memory, this was too large. Tables 15 and 16 show the results of the two chessmasters and the message lengths for each strategy. The results showed that the prime strategy used was the two ply strategy, this could be because a chess master may plan a form of attack that takes a few moves and not necessarily involves quiescence during that period. The quiescence search is unable to recognize the possible future attacks that might occur because of the current move's lack of quiescence. This theory is supported by the one ply's better inference than that of some of the quiescent models. The inability to distinguish certain combinations and putting emphasis on wrong moves, is demonstrated by the four ply model's poor performance in both cases. Another reason is highlighted by one of the cases in the analysis above that shows the two ply brute force is more commonly differentiated than that of the quiescent models.

 

Table 15: Inference results from inferring search strategies of Fischer.

	Inferred Strategy
Strategy	$\lambda_{MAT}$	$\lambda_{MOB}$	$\lambda_{CENT}$	$\lambda_{ATT}$	I-Inferred
hline min1max1	0.5	0	0.313	1.188	224.313
min1max2	0.625	0	0.188	0.938	225.548
min1max3	0.625	0	0.563	1.812	216.928
min1max4	0.75	-1	1.938	1.813	234.728
min2max2	1.125	0	0.688	1.188	199.631

 

Table 16: Inference results from inferring search strategies of Kasparov.

	Inferred Strategy
Strategy	$\lambda_{MAT}$	$\lambda_{MOB}$	$\lambda_{CENT}$	$\lambda_{ATT}$	I-Inferred
min1max1	0.313	0	0.313	1.125	278.172
min1max2	0.25	0	0.313	0.813	284.609
min1max3	0.314	0	0.438	0.813	282.3
min1max4	0.313	0	0.375	0.813	283.54
min2max2	1.437	0	0.438	0	261.697