Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Interesting result from SSDF

Author: blass uri

Date: 23:03:35 06/06/00

Go up one level in this thread


On June 07, 2000 at 00:34:01, Jorge wrote:

>On June 06, 2000 at 15:16:03, blass uri wrote:
>
>>On June 06, 2000 at 14:20:18, Jorge wrote:
>>
>>>On June 06, 2000 at 01:23:34, Ralf Elvsén wrote:
>>>
>>>>On June 06, 2000 at 01:13:49, Jouni Uski wrote:
>>>>
>>>>>On June 05, 2000 at 21:58:33, Robert Hyatt wrote:
>>>>>
>>>>>>On June 05, 2000 at 21:10:03, Wayne Lowrance wrote:
>>>>>>
>>>>>>>On June 05, 2000 at 02:39:35, blass uri wrote:
>>>>>>>
>>>>>>>>On June 05, 2000 at 02:33:01, Jeremiah Penery wrote:
>>>>>>>>
>>>>>>>>>On June 05, 2000 at 02:17:57, Jouni Uski wrote:
>>>>>>>>>
>>>>>>>>>>Look at this:
>>>>>>>>>>
>>>>>>>>>>Junior 6      1111½0½111101½1½11½1½1½111101101111111111 33.5/41
>>>>>>>>>>Crafty 17.07  0000½1½000010½0½00½0½0½000010010000000000  7.5/41
>>>>>>>>>>Note 10 lost games in row. This result indicates 260 rating points difference.
>>>>>>>>>>May be SSDF should update to 17.10 soon!
>>>>>>>>>>
>>>>>>>>>>Jouni
>>>>>>>>>
>>>>>>>>>I thought they did.  As far as I ever saw in the posted SSDF games/results,
>>>>>>>>>Junior6 was never very much ahead of Crafty in the match...
>>>>>>>>
>>>>>>>>The games that were posted were between Fritz6 and Crafty and not between
>>>>>>>>Junior6 and Crafty and it was also with Crafty17.07
>>>>>>>>
>>>>>>>>Uri
>>>>>>>This does not make sense. Crafty hangs tough with Fritz, Junior kills crafty and
>>>>>>>Junior and fritz are very close. With that many games played I would not expect
>>>>>>>to see that result for crafty. Crafty is a hell of a program and Junior is not
>>>>>>>that superior unless played at fast time controls where crafty falls short of
>>>>>>>the commercials (that evaluation is mine, Dr Bob may disagree).
>>>>>>>Wayne
>>>>>>
>>>>>>No.  I have always agreed that it is not doing as well in blitz as it is doing
>>>>>>at longer time controls.
>>>>>>
>>>>>>I am concerned about a 10-0 run, because I simply don't see nor expect that
>>>>>>against my program, or against any other program.  It suggests that something
>>>>>>has either gotten corrupted, the opening is not being controlled by book
>>>>>>learning, or something else.  I'll try to look when I have time, but it sure
>>>>>>looks odd...
>>>>>>
>>>>>>And then again, it could be perfectly correct for all I know.
>>>>>
>>>>>I made small calculation: In SSDF Junior is 65 points better than Crafty, so it
>>>>>can except 0.592 points/game. This means that probability for 10 in row is
>>>>>    10
>>>>>0.592    = 0.005 i.e. 0,5% so it's not impossible, but happens only once in
>>>>>190 try.
>>>>>
>>>>>Jouni
>>>>
>>>>The number 0.592 is based on wins and draws. If we have (say) 30% of
>>>>the games ending in draws, the number should be changed to
>>>>0.592 - 0.3*1/2 = 0.442, and 0.442^10 = 0.00028. No big deal though.
>>>>
>>>>Ralf
>>>
>>>Ralf, I'm trying to follow what you are saying about the probabilities, but
>>>something is wrong here. If 0.592 is approx. true for both wins and draws, then
>>>the P(Junior does not loose in 10 or more) much < .005, actually <
>>>0.005*2=0.000025 (since it happened twice in one event) which is a big deal!
>>
>>0.005*2=0.01 so I understand that you mean 0.005*0.005 and I use 0.005^2 for
>>this number.
>>I do not understand how do you get 0.005^2
>>
>>0.442 is the probability for Junior to win one game(assuming 30% draws and
>>expected result 0.592 in one game) so 0.442^10=0.00028 is the probability to win
>>10 games in a row.
>>
>>This assumption ignores the fact that the probability to win with white is not
>>the same as the probability to win with black and it also ignores the fact that
>>the events are dependable because of learning.
>>
>>learning can make the probability bigger because it is possible that after
>>enough games one program is going to start to win every game by repeating the
>>same opening when the other side's book is not big enough to avoid repeating the
>>same opening.
>>
>>The probability is also bigger because the match is of 41 games so the
>>probability to win 10 games in a row in the match is bigger than the probability
>>to win 10:0 in a match of 10 games
>>Uri
>
>
>Yes, I'm assuming randomness and independence which is a big factor, however i
>believe the Prob. of W is about the same as prob. of B winning in the long run
>and that the variabilty is the same for both colors.  Assuming Ralph's original
>calculations is somewhat true, P(Junior=Win or Draw)=0.592, then the P(Junior
>Wins or Drws =37 out of n=41 total games)= 41C37(.59)^37*(.41)^4= 0.0000095 (C
>is Combination), which means this event approximately happens 1 in 105,000
>trials! But listen, you dont need to calculate anything to "sense" there is
>something wrong here with the above games, which weighs heavily into Junior's
>favor.

I do not know if there is something wrong with the games.
There is a mistake in your calculations.

The assumption P(Junior=win or draw)=0.592 is wrong.
The assumption was that o.592 is the expected result in one game and not
p(Junior wins or draw)

If p(Junior draw)=0.3 then P(Junior wins)=0.442 then
p(Junior wins or draws)=0.742 by this assumption.

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.