Author: Uri Blass
Date: 10:00:20 02/12/03
Go up one level in this thread
On February 12, 2003 at 12:24:31, Andreas Herrmann wrote: >On February 12, 2003 at 09:30:17, Uri Blass wrote: > >>On February 12, 2003 at 08:30:55, Ed Schröder wrote: >> >>>On February 12, 2003 at 06:49:44, Thorsten Czub wrote: >>> >>>>The new Rebel12 BETA5 seems to be very strong. >>>>In games against Rebel XP REBEL.ENG >>>>this new effort of Ed Schröder played unbelievable >>>>12.5-0.5 !!!!! 12/1/0 (13) 96.2% on at 1' per move. >>> >>>It's strange indeed but still within the error margin. I once had a 14-1 case, >>>in the end it became about equal. But of course cases like this are the big >>>exceptions. Don't be surprised when you are at 50-100 games you are back at 60%. >> >>12.5-.5 or 14-1 are clearly significant results. >> >>I do not guess that the difference is so big like the result suggest but I will >>be surprised if they are almost equal. >> >>If I assume 40% for white 30% draws and 30% for black then I expect 12.5-.5 or >>13-0 to happen in clearly less than 0.01% of the cases. >>If I assume 7 games with white then I get by simple combinatoric >>0.4^7*0.3^6+ >>0.4^7*0.3^5*0.3*6+0.4^6*0.3^6*0.3*7 >> >>It is clearly less than 1 out of 10000 cases when the programs are equal. >> >>If I see 12.5-.5 or 14-1 then I am going to believe that one of the following >>happened: >> >>1)The machine is broken >>2)There is a big improvement and my guess is to expect more than 65% in more >>games(Note that I do not expect more than 90% because I believe that part of it >>has to be some luck). >> >>Thorsten found that 1 is false so I guess that 2 is correct. >> >>It is possible that with more information I will have a different opinion(for >>example if I know that the 12.5-.5 came only as result of a small change then I >>am not going to expect more than 65%) but when I do not know what you did and I >>see that result my guess is at least 65% for beta5. >> >>Uri > >i often let the absolutely same Holmes version play against itself under >Wínboard (2 times installed in different folders) to check only the stability. >Before about some weeks i had a result of 17,0 : 3,0. I have checked the >settings very carefully, and all was ok. So such results can happen. > >I can also remember results, where i have playd games between 2 different >versions of Holmes where one version scores 70% to 75% in 50 games. Then i have >continued, and after 150 games the score was only about 50%. That's real >statistic. > >Andreas I know statistics but 1)12.5-.5 is clearly more than 17-3 2)17-3 seems also strange between the same version against itself. I think that if everything is ok in your machine and the programs run on the same speed you need to play thousands of games before it happens. even with no draws 17-3 out of 20 games happen only in less than 3 out of 1000 matches and draws clearly increase the chances for smaller results. 1351/524288 is the exact proabibility to get 17-3,18-2,19-1 or 20-0 with 50% for both sides in every game Here we got 12.5-.5 and now 14.5-.5 in the first days. It does not make sense to guess that it is not a big improvement. 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.