Chessmaster games

Using the technique on a chessmaster game is the next logical step. The games are among the longest played, with the winning player and side displayed with a (W) White or (B) for Black. The Chessmasters chosen are the same as that in the Jansen et al.'s paper and the specific games chosen were:

• 11 games played by Kasparov in an It event in Moscow Russia in 1996
• 11 games played by Fischer in the 1963 USA Championships;
• Sherwin-Fischer, USA Championship, 1966, 100 moves;
• Fischer-Taimanov, Vancouver, 1971, 89 moves;
• Kamsky-Kasparov, New York, 1989, 107 moves;
• Kasparov-Karpov, 1990, 102 moves;
• Kasparov-Vladimir, Intel Chess Grand Prix (blitz), Paris (France), 1994, 86 moves (Note: This is not done in Jansen et al.).

The results for the inference on the chessmaster games for the advanced model as shown in Table 4 did not match that of Jansen et al.'s. Due, most likely, to the complexities involved in the creation of the chess program and the inference programs. The results are, however, similar, this discrepancy could be a result of numerical accuracy, search method, local vs. global minima or another probable cause, but the results for the chessmasters must be around these inferred values. It could be assumed that the differences in the implementation have caused this result, or perhaps even worse, that one of the databases used did not have the correct games recorded. It was deemed unnecessary to spend a few months looking for the source of these differences, as the general outcomes of the two papers are close. The results obtained, although not as accurate in predictions as Jansen et al.'s are still better than that of the one ply model. This would indicate that the higher search depth and more attributes are a better model of the chessmasters style. The results show that Kasparov was slightly better modelled by this technique than Fischer. The Kasparov model is again compressed more than that of the Fischer model, which shows that this result before was accurate.

 

Table 4: Inference results from chessmaster games using the advanced model.

	No.
Player	Moves	$\lambda_{MAT}$	$\lambda_{MOB}$	$\lambda_{CENT}$	$\lambda_{ATT}$	I-Random	I-Inferred	Comp (%)
Kasparov (J)	102(W)	1.20	0.09	0.10	3.19	481.4	370.8	$77.0\%$
Kasparov	102(W)	0.67	0.02	0.43	1.24	481.4	434.4	$90.2\%$
Kasparov (J)	107 (B)	0.82	0.18	0.09	1.81	443.7	328.7	$74.1\%$
Kasparov	107 (B)	0.57	0	0.53	1.62	498.6	438.1	$87.8\%$
Kasparov	86 (W)	0.63	0.10	0.66	1.21	357.3	292.1	$81.7\%$
Fischer (J)	89(W)	1.31	0.00	0.06	0.89	385.5	307.6	$79.8\%$
Fischer	89(W)	0.98	0	0.37	1.09	386.2	340.3	$88.1\%$
Fischer (J)	100(B)	1.61	-0.06	0.09	2.82	416.7	339.1	$81.4\%$
Fischer	100(B)	0.58	0	0.89	0.59	421.2	376.7	$89.3\%$
Fischer (J)	11 Games	1.24	0.04	0.09	1.15	1995.3	1555.3	$77.9\%$
Fischer	11 Games	0.5	0.02	0.48	1.34	1976.0	1747.0	$88.4\%$
Kasparov	11 Games	0.67	0.04	0.13	1.20	1387.88	1253.15	$86.7\%$
Actual Values		7.0	0.2	0.5	3.0