Author: Vincent Diepeveen
Date: 22:52:32 01/28/00
Go up one level in this thread
On January 29, 2000 at 00:47:58, Christophe Theron wrote: >On January 28, 2000 at 13:19:20, blass uri wrote: > >>On January 28, 2000 at 12:31:29, Shep wrote: >> >>>On January 28, 2000 at 11:31:29, Amir Ban wrote: >>> >>>>On January 28, 2000 at 05:25:26, Shep wrote: >>>> >>>>>On January 27, 2000 at 10:40:05, Jari Huikari wrote: >>>>> >>>>>>Where you consider the position of a game changes to middle game / end game? >>>>>> >>>>>>opening == most pieces haven't moved yet ? / opening book not ended yet ? >>>>>>end game == few pieces left ? >>>>>>middle game == neither of the two above ? >>>>> >>>>>I suppose some programs have a more fine-grained approach to this. >>>>>Tiger for example divides the game into several "phases" (more than 3 for sure >>>>>:) according to the material on the board, and different evals >>>>>(/extensions/pruning/...?) are fired off according to the phase the program is >>>>>in. >>>>>In the DOS version, it would show "Phase x/y"; in Rebel Tiger, the best way to >>>>>see it is when the eval suddenly jumps from like "+1.10" to "+0.20" (or vice >>>>>versa) after a capture has occurred. (Contrary to Fritz, this is _not_ due to >>>>>any preprocessor oddities. :) >>>>> >>>> >>>>Why isn't it like Fritz ? It sounds exactly the same. >>> >>>It does not happen to Tiger that he says "+0.00" and then after the next >>>capture, shows "-3.00" immediately. That would be a result that has been >>>reported repeatedly about Fritz. Tiger's eval may go up or down a bit, but it >>>does not miss losing moves just because it was "in the wrong phase". >>>So I suppose there's a difference between the preprocessing Fritz employs and >>>the things Tiger does between phases. >>> >>>Of course I cannot point the finger to it, not knowing either program's source >>>code, but I figure that Tiger's approach is different (besides, such drastic >>>eval changes (1 pawn or more) are _extremely_ rare for Tiger). >> >> >>I do not understand. >> >>Do you mean to say that the only difference between tiger and fritz is in the >>size of the change in the evaluation? >> >>Can tiger show scores of +0.4,+0.5,+0.5 +0.46 and never showing +0.9 and after >>the expected moves scores of +0.9,+0.95,+0.93,+0.96? >> >>If it can than I see no difference between tiger and fritz (except the size of >>the change) >> >>It is possible that you meant that tiger can show something like >> >>depth 7 +0.4 >>depth 8 +0.5 >>depth 9 +0.5 >>depth 10-13 +0.9 >> >>and after the expected move >>it can show from depth 1 to depth 10 evaluations of +0.9? >> >>If this is the case than tiger is not a root processor but can have jumps in the >>evaluation because of being a processor of something that is not the root but >>close to the root. >> >>Uri > > >You cannot use this to judge if a program is a root processor or not. >The score of a deep line can be stored in the hash table, and once it is found >in one search (maybe after a very long time), it can be found immediately in the >next search because of HT persistency. > the score changes for some part sure can be from hashtable, but not in this drastic way Christophe! This is like Ed saying he doesn't use the nullmove idea! You can change the type of car, but it remains a car! Tiger is obviously suffering from the fritz effect, after queen gets off suddenly score changes drastically. For example against DIEP in dutch open diep was at -1.0, tiger at +1.0. Tiger offers queen exchange. Diep wasn't expecting that and would not have done it in the same way. Diep happily exchanges and goes to very close to zero, before capturing it already goes to a score X which is near to zero. After physically moving the queen and exchanging it, my score doesn't change much, though hashtable is quite some bigger and having more probes. Score difference is 0.00 usually when relatively seen (11 ply X score goes to after making 2 plies at board to about X score at 9 ply). Tiger dropped nearly a pawn DIRECTLY at very small depths already after diep took the queen. That is not a hashtable issue! Tiger changes incredible much. doesn't take away that you can evaluate a lot of other things in positions, but preprocessor obviously is very important to Tiger's root score! I saw at dutch open that tiger uses a kind of gnuchess stage (0..9) to express where it was, and it putted that in the root to the screen. I don't understand why you print out a GNUchess stage at the root Christophe at the screen! For a preprocessor it is needed though. > Christophe Vincent
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.