Author: Uri Blass
Date: 14:11:02 12/28/03
Go up one level in this thread
On December 28, 2003 at 16:39:01, Vincent Diepeveen wrote: >On December 28, 2003 at 16:25:06, Uri Blass wrote: > >>On December 28, 2003 at 15:58:40, Vincent Diepeveen wrote: >> >>>On December 27, 2003 at 14:05:23, Robert Hyatt wrote: >>> >>>>On December 27, 2003 at 12:36:32, Mridul Muralidharan wrote: >>>> >>>>>On December 27, 2003 at 09:04:05, Sune Fischer wrote: >>>>> >>>>>>On December 27, 2003 at 04:58:51, Mridul Muralidharan wrote: >>>>>> >>>>>>>4) Even with a 1:10 or 1:8 advantage crafty only barely manages to catch up or >>>>>>>beat these top order programs - so not much of a chance if they show up with the >>>>>>>above mentioned machines. >>>>>> >>>>>>I believe you are jumping to conclusions with "not much of a chance.." :) >>>>>> >>>>>>Crafty seems to be more than just barely holding on, it seems to be winning >>>>>>these matches at 8-10x speed advantage. >>>>>>That would mean Crafty still has chances even with "only" a 4x speed advantage. >>>>>> >>>>>>I'm not saying it would be a favorite but it's enough of a lottery to say that >>>>>>there are chances, and who knows what kind of hardware Bob could get his hands >>>>>>on ;) >>>>>> >>>>>>-S. >>>>>>>Mridul >>>>> >>>>>True - a good opening line which is favourale to crafty - and it could win even >>>>>with a 1 : 2 disadvantage :) >>>>>But it should be a good line - and this is like a jackpot - we cant count on >>>>>this ;) >>>>>On a even position - with lots of possibilities (not one of those dead drawn >>>>>boring opening lines) - it would be tougher. >>>>>If 1:8 just barely helps - then 1:4 is going to be a real challenge - also note >>>>>- it is going to be lower than 1:4 - not even 1:4 >>>>> >>>>>Mridul >>>> >>>>That's wrong. Pick your current box. I can _certainly_ find something that >>>>will be 4x faster than any box that showed up at the WCCC this year. And that >>> >>>But nothing where crafty runs at, so i laugh loud for this! >>> >>>You cry some about a few minute tests you were allowed to do at a 32 processor >>>shared memory bus Alpha. Show the logs, no one is believing you here! >>> >>>Note the crafty-junior match, i score at home 95% against junior in tests equal >>>hardware, you have problems getting 50% against an outdated PII. >> >>I do not believe that you score at home 95% at equal hardware. > >I'm not the only one scoring that against junior with tournament book. 95%? I do not believe it. > >> >>You even never proved that you can score even 50% against Crafty on equal >>hardware. > >?? > >I kill it here completely. > >*completely*. I never saw a proof for it. Can you prove it in a match against Crafty on ICC? > >Level played is usually 2 hours + 30 seconds increment a move and in other cases >40 in 2 + 1 all > >>I do not see Diep in the ssdf list. > >I need to pay them 2 machines to garantuee that i get on the list. >I refuse to pay 2 machines to garantuee that diep gets quick on the list. I do not know about a rule that you need to pay 2 machines. If it is only to be in the list in few months and not in 1 year then it is still be better for you to have diep of today in the list so at least you can prove that it is better than the programs of today. > >Further if you play under fritz8 it will eat 80% system time and your engine >getting 20%, additionally give commands like putting your hashtable to 1 MB. I doubt if you are right but you do not need to play under fritz8. I think that you can play under arena. > >So basically you are busy 6 months just avoiding to lose because of details they >have set favourable for them. > >SSDF is a paid list and even paying $10 i would not mind, but for a paid list >you expect that *all* games they play get posted, not 50% of the games. > >Further is the human interpretation in testing always the most important, i >remember a tester mine a few years ago who concluded diep got butchered by Rebel >at the time and only a few draws. > >However those 'draws' were wins for diep. When a rook up in endgame it arbitered >it still as a draw the rebel interface. > >So you see all companies have made endless tricks for SSDF and you need to pay a >shitload of money too and for what? > >The only thing that counts is world champs anyway. Arguably IPCCC tournament is >important too because of a serious level. I do not see why you think it is the only tournament that is important. I consider WBEC as an important tournament and some commercial programs like Ruffian2 and The king are participating. > >>If you can get even 70% against Junior8 and other programs of the same level >>then I suggest that you prove it in the ssdf games. > >I do not join events which are paid and unclear. SSDF is a paid list where you >must avoid 100 traps first and pay to garantuee that you get at the next list at >the list. What about WBEC? I pay nothing and all the games are available. time control is only 40/40 that is not enough for you but hardware get faster every year. What about thorsten's tournament? I remember that Diep played there in the past and after failure it did not play again. Movei is playing there and so far has a good result of 1.5/3 including a win with black against Fritz8(I do not claim that today it is close to the level of Fritz8 but in one game everything can happen and Fritz8 chose a bad line and after sacrificing a pawn out of book it sacrificed another pawn and got no compensation so movei traded all the pieces and got winning KRPP vs KR) > >If there is a delay of 3 months or so then trivially they can ship new books and >kill you again. > >>It may be possible with a lot of time to prepare some killer book against >>something that has a known book(you simply play against a copy of the program >>and even if you win only 100 games out of 1000 you can add the opening of the >>100 games that you won into your book) but Crafty did not win by this method. > >>Diep lost against the top programs in the world champpionship because of your >>evaluation bugs and not because of a bad book. > >Book was ok in world champs. Very ok. > >Hardware was ok too. > >You see the proof is so easy that search depth is not holy. Some old stubborn >persons here believe it is not true. Fixing bugs in evaluation is the only thing >that counts. 2000 lines isn't enough for that! No I do not see the proof. I always said that bigger is not necessary better and bigger may be worse because of bugs. > >>> >>>However when they show up at world champs they use quite a better book... ...try >>>to beat that i'd say! >>> >>>So show up at world champs instead of spitting all this crap here. Everyone i >>>knows wins from other programs when they test at home. >> >>Not everyone. >> >>Did GreenLight win from other programs when the programmer tested at home? > >>Did Falcon win from other programs when the programmer tested at home(it did >>good results but it is clear that his results were not when he gave the >>opponents the hardware advantage that they have in WCCC). > >>I do not know if Crafty could appear with 10:1 hardware advantage but with that >>advantage I do not see a proof that it is not competitive. > >You know just as well that crafty doesn't run at machines with more than 8 cpu's >which are cc-NUMA, so it will never run at a supercomputer and never have a 10 >to 1 advantage in hardware. I do not know. > >Simply because it can not run at supercomputers of course. > >>The only claim that you can make is that the time control is not slow enough and >>people should use slower time control to have better simulation of what could >>happen in WCCC. >>Uri > >Even falcon scores at home 60% against Junior8. Falcon did with Fritz8.ctg when Junior also used Fritz8.ctg. Crafty played its match with its own book when commercial programs also used their own book. > >Note i tested against deepjunior8 basically i know they show up at a quad xeon >at world champs so testing at a fast level is bullshit. > >In tests you must take into account what hardware people show up with! > >Diep gets equal hardware then here. > >But the proof is trivial that only evaluation matters, or do you disagree still? > >Best regards, >Vincent I disagree. Not only evaluation matters. The same evaluation with a machine that is 10 times faster get clearly better results against the same opponents. Uri
This page took 0.01 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.