Author: Frank Schubert
Date: 03:31:01 03/01/00
Go up one level in this thread
On February 29, 2000 at 13:20:35, Masciulli Gianluigi wrote: >On February 29, 2000 at 12:48:07, Tony Hedlund wrote: > >>On February 29, 2000 at 09:52:06, Masciulli Gianluigi wrote: >> >>tscp15's 90.2% against 1700 = 2062 >>fortress161's 96.9% against 1700 = 2232 >>golem01's 0% against 1700 is < 1000 >>fortress16's 58.8% against tscp15 indicates a difference with 64 points, >>2179-2115. > >thanks tony for your answer. > >ok ... that's the easy part but ... >you see you listed frotress TWICE with perf. 2232 and 2179-2115. >I need the global performance. > >> >>>One more: I'd like to have extimes about probability that performances match the >>>true elo on the base of games played (ssdf Like)... any matematician out there? >> >>This question I do not understand, sorry. > >I'll try to explay this better: > >if I play 1 billion game and the result between prog.A (elo ??) and Prog.B .. >prog.C prog.D (and all have elo 2000) is esactly 50% >I can say that the elo of prog A in that set of programs is almost esactly >2000. >but what about the same result after only 50 games. >may be there is some difference between elo and perf. in this case. >may be i can say, in this case that the elo of the program A is 2000 with an >error margin of 50 point (in other words there is probability .9 that the true >elo is within the range 1950 -> 2050). >I want to calc the error margin. >I hope is clear now. > >gianluigi Hi, in a few days you can download my program ELOStat which was used to calculate the statistics of the CadaquƩs tourney (look at http://www.computerschach.de/tourn/frank01.htm) ELOStat also gives the error margins you mentioned. I will post ist here, when I am ready. Frank
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.