|
AW: Re: Die tollen Schachprogramme von Arne Bailliere
Hi Nick
If the baseline is set very high like 2800 elo, the search depth increases , so this will have an influence of the general strength of the players because the tresholds at this level become very narrow. But the lower you set the ELO rating , the more influential the style and plan mechanisms become since they crystallize knowledge about the future possibilites or goals based on sequences, patterns and the plan extraction from backward plan analysis we performed on all of the master games. So the lower the elo set , the more the plan and style rules start to influence play. So in Tournament mode , you can organise tournaments with a 1600 elo Capablanca vs a 1800 elo Karpov and 2000 elo Smyslov and so on . Basically you can adapt the personality to any level, and it should still retain its playing style , maybe not just too accurate anymore at 1200 Elo. But it will still try to boost moves that fit plans , patterns and sequences because the system "knows", or tries to know, that these moves lead to a possible advantage down the road. Plans are held together by causal chains and these chains get a boost the longer they persist in the plan memory. If the position changes too much the plan is of course abandoned. They also have their own historic opening book from their own games. So far i have noticed the plan and style consistencies work well at lower levels for positional personalities like Smyslov and Capa, but not always so well for more pure calculation - sacrifial players like Tal and Kasparov - in some cases. But sometimes it depends also on the local installation, since at Thorsten´s place The Chessmind Tal really made some incredible sacrificial attacks that I have not seen myself yet on my system! I have had various results on different levels of Tournaments. Capablanca is always performing strong as well as Karpov and Smyslov. Kasparov and Fischer close behind, and Tal with occasional beauties because at shallow depths the attacks and sacrifices sometimes get refuted. Tal works better vs humans on the system istnalled on my pc .
|