Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Junior - Crafty NPS Challenge - a user experiment

Author: Robert Hyatt

Date: 18:18:26 11/22/03

Go up one level in this thread


On November 22, 2003 at 16:26:01, Uri Blass wrote:

>On November 22, 2003 at 15:56:08, Robert Hyatt wrote:
>
>>On November 22, 2003 at 01:01:35, Peter Kappler wrote:
>>
>>>On November 21, 2003 at 22:03:53, Robert Hyatt wrote:
>>>
>>>>On November 21, 2003 at 19:11:41, Will Singleton wrote:
>>>>
>>>>>On November 21, 2003 at 18:24:42, Robert Hyatt wrote:
>>>>>
>>>>>>On November 21, 2003 at 18:05:55, Joachim Rang wrote:
>>>>>>
>>>>>>>On November 21, 2003 at 16:47:52, Peter Berger wrote:
>>>>>>>
>>>>>>>>Following recent heated discussions I'd love to do a little testmatch. There are
>>>>>>>>different versions of the challenge online - I chose one that I can kind of
>>>>>>>>simulate myself with slower hardware.
>>>>>>>>
>>>>>>>>Junior 8.0.0.2 will play on a P233MMX, 32 MB RAM.
>>>>>>>>Crafty 19.4 will play on a PIV2.0GHz notebook, 1GHz RAM.
>>>>>>>>
>>>>>>>>Time control will be game in 2 hours with 10 seconds increment/move. The match
>>>>>>>>will be done like older FIDE world championship matches - the first one to win 6
>>>>>>>>games wins the match, draws won't count.
>>>>>>>>
>>>>>>>>Junior uses 16MB Hash, 3+4 men tablebases, 1MB cache, junior8.ctg.
>>>>>>>>Crafty uses 384MB Hash, 64MB hashp, 3+4 men tablebases, 32MB cache, own book,
>>>>>>>>aware of playing a computer.
>>>>>>>>
>>>>>>>>Compairing setups with Crafty bench (hash 12M, hashp 3M, cache 1M on the slower
>>>>>>>>one) suggests a speed difference factor of about 10.5 in raw nodes per seconds
>>>>>>>>and 11.0 in "SMP time to-ply-measurement" between the two computers.
>>>>>>>>As the Junior Mark doesn't work on the slower one and Junior chooses to search
>>>>>>>>different depths on both in the starting position, I can't really give a number.
>>>>>>>>The difference seems to be slightly lower for Junior though, sth like 9.0 maybe.
>>>>>>>>
>>>>>>>>Saying that the faster computer is about 10 times faster shouldn't be too wrong.
>>>>>>>>
>>>>>>>>That's also clearly an upper-bound for faster hardware Crafty could reasonably
>>>>>>>>come up to compete with against a single-CPU opponent in a current competition
>>>>>>>>on fast computers IMHO - the speedup demands 16 CPUs I guess, and I don't know
>>>>>>>>if Crafty can really scale that well.
>>>>>>>>
>>>>>>>>With this setup Crafty should be the clear favourite I suppose.
>>>>>>>>
>>>>>>>>Crafty won the toss and will have the white pieces in the first game.
>>>>>>>>
>>>>>>>>Peter
>>>>>>>
>>>>>>>
>>>>>>>nice test indeed. I think crafty will win.
>>>>>>>
>>>>>>>regards Joachim
>>>>>>
>>>>>>If it loses I should obviously retire from computer chess due to gross
>>>>>>incompetence.
>>>>>
>>>>>You already have two people (above) who think Junior has a chance to win, so I
>>>>>suppose you should at least start checking the social security balance.  Me, I
>>>>>think Crafty will win easily, perhaps 6-0 (not counting draws).
>>>>>
>>>>>Will
>>>>
>>>>I'd expect 5-1 or 6-0 at 10:1 time odds, discounting quite a few draws
>>>>that are expected...
>>>>
>>>>As for who thinks that a 10:1 time odds is not enough, I suspect they have
>>>>not "gotten dirty" by writing something that is serious.  10:1 is quite a
>>>>bonus.  100:1 is even better...
>>>>
>>>>If I couldn't win at 10:1 it would be time to hang it up, IMHO, no matter
>>>>how much or little time I get to spend on it...
>>>>
>>>
>>>
>>>When did the time odds change from 100x to 10x?
>>
>>There are _two_ things going on, which is causing confusion.  Amir and I
>>are talking about 100:1 time odds.  The match being played on CCC right
>>now is using a 10:1 handicap by using a slow pentium 233 vs a new machine
>>that is about 10x faster.
>>
>>This is just a "proof of concept" to see if 100:1 will be significant or
>>not.
>>
>>
>>>
>>>The SSDF rating list shows Junior ahead by 25 ELO even when Crafty runs on a
>>>machine that is 4x faster.
>>>
>>>Junior 7.0      128MB K6-2 450 MHz      2639
>>>Crafty 18.12/CB 256MB Athlon 1200 MHz 	2615
>>>
>>>(My Athlon 1400 is 4.5x faster than my K6-3/450, so I think 4x is about right
>>>for the two machines above.)
>>>
>>>The same rating list shows that Junior 8.0 is 100 pts stronger than Junior 7.  I
>>>don't know how much progress Crafty has made since version 18.12, but I would
>>>guess that it's less than 100 pts.
>>>
>>>You'll also likely be facing Junior's tournament opening book - potentially a
>>>huge advantage for Junior.
>>
>>No, the book is _specifically_ eliminated.  This is testing speed vs speed,
>>not book vs book which is uninteresting in time odds.  If you saw my suggestion
>>for rules, "if either program is -1 or worse out of book, the game is aborted
>>and a new one started with the same colors, to eliminate book traps that should
>>not matter when comparing two engines at time odds."
>
>book traps are not eliminated in this way because it is possible to get position
>that both programs do not understand but can find the right moves because it is
>clear that there is no good alternative and only some moves after they are out
>of book one program understand it is -1 worse.

Not much I can do about this, except that if both think it is pretty even,
then a deep trap is just something that happens.  However, if it is a sac that
neither program understands, then one will be +1 and the other will be -1, and
that's enough to kick the opening out and start over.


>
>The only way to eliminate book is to play something similiar to the nunn match
>or in other words after a game is finished to play the same opening with
>opposite colors.

Or, play an opening, then play the same opening with colors reversed...  Of
course one could have more book than the other, but that's ok...




>
>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.