Author: Steffen Jakob
Date: 00:15:36 04/30/02
Go up one level in this thread
On April 29, 2002 at 16:06:31, José Carlos wrote: >On April 29, 2002 at 15:52:32, Scott Gasch wrote: > >>On April 29, 2002 at 14:01:23, Joe McCarro wrote: >> >>>If I were playing someone over the board and they seemed to give me a >>>possibility to play Bxa1 snatching the rook I would think long and hard before >>>doing that. I'd figure as long as this isn't a trap I will win the game. Let me >>>take my time to just make sure its not a trap. I wonder if this couldn't be >>>programmed in. Anytime the other player makes what on the surface appears to be >>>a blunder (e.g., drops over a pawn) the computer could spend extra time working >>>out the position before moving. If it ended up it was in fact just a blunder >>>presumably the computer should still be able to win despite the extra time spent >>>looking for the tactical shot. If it found it wan't a blunder the computer >>>might avoid taking the poison. Do the programmers do anything like this? Would >>>this in fact be helpful or would it have disadvantages as well? >> >>This idea could certainly be programmed. But there's a problem. Let's assume >>the engine is playing a blitz game and has allocated itself 10 seconds to think >>about a move. In 10 seconds it gets to depth d and realizes that the move it >>wants to make seems to "win". To get to depth d+1 will a long time, maybe 10 >>more seconds or so. To get to depth d+2 may take 10 + 20 more seconds >>(sometimes more). >> >>So my point is, to say "this position is important, I should think about it a >>bit more" is fine... but to think about it even just a little bit more may take >>a really long time. The question is whether its a good use of the engine's >>allotted time. Maybe. >> >>Personally, I do some tricks with time allocation but nothing as direct as the >>idea you suggest. I consider it just too expensive, especially at fast time >>controls. >> >>Scott > > Some time ago I (Averno) was playing Hossa on ICC. Hossa took a long time to >reply a quite normal move and Steffen told me Hossa extended time when there was >a big jump (up or down) in the score, exactly with the above stated idea. > I'm not quite sure about it, but it'd be certainly interesting if Steffen >himself could tell us his experience with that time extension. Hossa is a slow searcher. Most of the time he gets outsearched against other comps. Therefore it doesnt happen very often that a computer opponent blunders away material which Hossa sees very quickly. Instead it is likely that in a deeper search the score will drop and also playing the obvious move from the shallow search would lose. Therefore Hossa extends the time there. Of course I consider how much time I have overall, if there is an increment, how big the score increment was etc. The idea is simple: 1. if it was really a blunder then I will win anyway. 2. If it was a trap-like move then thinking deeper might safe my life (which happens rather often!). I think adding this was an improvement for Hossa. This might not be true for other engines which dont get outsearched as often as Hossa. Best wishes, Steffen.
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.