Author: Uri Blass
Date: 08:37:54 02/25/04
Go up one level in this thread
On February 25, 2004 at 11:25:50, martin fierz wrote: >On February 25, 2004 at 10:58:36, Robert Hyatt wrote: > >>On February 25, 2004 at 09:13:43, martin fierz wrote: >> >>>On February 25, 2004 at 07:02:11, Dieter Buerssner wrote: >>> >>>>On February 25, 2004 at 05:56:16, martin fierz wrote: >>>> >>>>>it won't pop *my* eyes. i once reduced hash key sizes in my checkers program >>>>>beyond all sensible settings, because there was a discussion here about whether >>>>>you really need 64-bit keys. in my checkers program, i have 64 bit keys, but >>>>>effectively it's only using about 52 bits. i have about a 20 bit part which is >>>>>used for the hashindex with %, and of the remaining 44 bits i store only 32 as a >>>>>check. i reduced those 32 down to about 8 (!!) bits and in 100 test positions >>>>>only saw one different move played IIRC. ridiculous, i must have lots of >>>>>collisions there. unfortunately, i didn't count the collision number, or write >>>>>down the results - but i know what you're talking about! >>>> >>>>Almost the same experiment with my chess engine (inluding many details, like the >>>>effective number of bits used, and going down to 8 bits only): >>>>http://chessprogramming.org/cccsearch/ccc.php?art_id=190318 >>>> >>>>Regards, >>>>Dieter >>> >>>hi dieter, >>> >>>i had forgotten about your post on this, but now i remember it. very similar to >>>my observations, and if only we had written our observations up a bit more >>>seriously we could have written the paper that bob is publishing now ;-) >>> >>>cheers >>> martin >> >> >>Hey, I'm easy to get along with here. :) >> >>I have already asked one other person to do some similar testing. I'd be happy >>to tell you both what I have done, and have you run similar tests, and join me >>as authors on this paper. >> >>I am doing the test slightly different, as rather than a specific number of >>signature bits, I am forcing a particular error rate (ie one error every N >>nodes) with the idea being that I should be able to choose N in 1 error every N >>nodes such that the score never changes, or the score changes or not the best >>move, or the best move changes but it is not a bad change, or the best move >>changes and it probably changes the game outcome. >> >>If either/both are interested, email me and I can send you a draft, which >>explains how I am testing, and includes the test positions I am using. I have >>some endgame positions (ie like fine 70), some sharp tactical positions like the >>Ba3 Botvinnik-Capablanca move, and some plain middlegame positions from games >>Crafty played on ICC. >> >>Let me know if you are interested... > >hi bob, > >this wasn't intended as criticism :-) >you are a computer scientist, and i am not; it is your job to write this sort of >paper - mine would be to write papers about physics... >anyway, i have nothing to contribute chess-wise: my program is 200-300 rating >points weaker than crafty, and i don't believe in writing academic papers about >"toy programs". as you recently pointed out to me, you do some things different >now that you search deeper (no more pin detection, higher R for nullmove). >for example, i could write a paper about how R=1 is better than R=2 for my toy >program, which is completely irrelevant in general, because for good programs >it's clear that R=2 is better than R=1. so the only thing i could contribute is >a similar experiment for checkers, where my program is much more advanced than >my chess program. but i doubt that that would really be interesting either :-) > >cheers > martin The last time that I checked R=3 was better than R=2 for my movei. I never tried R=1 but I am almost sure by intuition that if R=3 is better than R=2 then R=2 is better than R=1. I use R=3 without verified null move pruning in the middle game and R=3 with verified null move pruning in the endgame. 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.