Author: Ed Schröder
Date: 15:32:40 10/05/98
Go up one level in this thread
>>Hi Steffen, >>I don't use the piece-square tables to make Rebel play 2.c4 >>I have made a chess tree from a good database. Rebel consults the chess tree and >>finds for 2.c4 a good score (say 55%) in (say 4000) games with an average elo of >>(say) 2400. >>Based on this data I add a "flexible" bonus to 2.c4 >>The bonus can vary from -1.00 to +1.00 >>Rebel gives the following information: >>Move % Games Bonus >>2.c4 58% 17.806 +0.52 >>2.Nf3 55% 6.775 +0.23 >>2.e4 60% 379 +0.25 >>2.Bg5 59% 299 +0.25 >>So based on the chess tree a 0.52 bonus is added to 2.c4 which will force >>Rebel to play the move. >Ed I assume that the bonus is derived from some combination of the number of >games plus the score %. Can you tell us the exact equation? In full detail is too complicated but your above guess is quite right. In general it goes like this: - Base percentage WHITE is 55% - Base percentage BLACK is 45% - For each percentage below the base give a penalty. - For each percentage above the base give a bonus. - Then this value modified with a formula to ensure that moves which are played more often are in principal more reliable and this will reflect in the final bonus/penalty value for the move(s). 2.c4 58% 17.806 +0.52 2.Nf3 55% 6.775 +0.23 2.e4 60% 379 +0.25 2.Bg5 59% 299 +0.25 From the above example you can see that although 2.e4 has the highest percentage (60%) it is only rewarded with +0.25 while 2.c4 has just 58% but is rewarded with 0.52 This comes because 2.c4 has much more games played than 2.e4. The bottom line is that if 17.806 people have played 2.c4 and scored 58% this must be a good move which justifies the 1/2 pawn bonus. - Ed - >-- >Komputer Korner
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.