Author: Uri Blass
Date: 06:28:47 08/25/01
Go up one level in this thread
On August 25, 2001 at 00:33:23, Christophe Theron wrote: >On August 24, 2001 at 14:23:58, Uri Blass wrote: > >>On August 24, 2001 at 13:44:53, Christophe Theron wrote: >> >>>On August 24, 2001 at 13:33:33, Thorsten Czub wrote: >>> >>>>On August 24, 2001 at 12:33:26, Christophe Theron wrote: >>>> >>>>>On August 24, 2001 at 11:41:47, Harald Faber wrote: >>>>> >>>>>>With these losses it is justified that Tiger didn't become single world >>>>>>champion. What is your explanation for this result? Did you play GT in >>>>>>aggressive mode? What made Tiger lose especially the last game vs. an amateur? >>>>> >>>>> >>>>> >>>>>I can answer your question if you answer mine: >>>>> >>>>>When you flip a coin, why is it head sometimes, and some other time it is tail? >>>>> >>>>> >>>>> Christophe >>>> >>>>i don't think this is an explanation christophe. >>>>maybe we should look into the games instead. >>> >>> >>> >>>It is an explanation. >>> >>>If your program is able to win 60% of the games against a given opponent, you >>>are not going to consider to rewrite it every time you see one of the 40% of >>>losses. >>> >>>If you only look at the losses you are going to believe that the program is shit >>>and start doing a lot of dramatic changes. Probably you are going to fix some of >>>the losses, but it is very likely that you are also change the outcome of a >>>significant part of the 60% of games that were won. >>> >>>In the end you'll be very satisfied because the program does not lose the >>>previously lost games anymore, but oops, it does not win anymore the ones it had >>>won before! >>> >>>So I'll do as I always do after a big tournament: I'll NOT study closely the >>>lost games (unless I can see a gross mistake by quickly browsing thru the games) >>>and keep on working as I do normally. >> >>I think that understanding the reason that the program lost may be important. >>It does not mean to do changes without testing them but knowing the reason that >>the program lost may give you good ideas for changes. >> >>Some reasons are tactical mistakes. >>I did not look at the game against shredder. >>The losses against Junior and Sos seems to be because of tactical mistakes >>and it may be interesting to investigate them in order to know more(I did not do >>it) > > >If the reasons are tactical, then I have even more reasons to not care about >these games. I already spend a lot of time working on the tactical abilities of >Tiger, so the problems (if there is any problem) can be solved by themselves >when my tactical algorithms are improved. I read from nemeth's post that 20.Bb3+ was the losing error against Persos. My Gambittiger liked e5 in the first 13 plies. It changes it's mind to 20.Bb3+ at depth 14 and changes later it's mind to e5 at depth 15. The score of 20.Bb3+ dropped from +0.54 at depth 14 when tiger finds Bb3+ to -0.20 at depth 15. Tiger finds 20.e5 with score of +0.10 at depth 15 Gambittiger2 needs 17 minutes on PIII850(184 Mbytes hash) to see 20.e5. I told it time control of 400 minutes per 40 moves that means practically 400 minutes for moves 20-40 and it continues to search depth 16 and the score for e5 after 31:17 is 0.30(I do not post the all analysis because I use my p200 to post when my PIII850 is not connected to the internet). It seems that nemeth is right that 20.Bb3+ is the losing error. It will be interesting to know the depth, the score and the time of Tiger14.6 in the game. Maybe a better time management could help Gambit to find 20.e5 against Persos and maybe equal hardware could help tiger. > >If it is because the tactics were out of the horizon of any program, then it's >just bad luck, and I do not see any reason to worry about it. > >I would prefer evaluations reasons, because in this case it is possible to add >specific knowledge in the program in order to fix it. specific knowledge in the program can be about search and not only about the evaluation. Uri
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.