Author: Dann Corbit
Date: 18:09:36 04/30/01
Go up one level in this thread
On April 30, 2001 at 19:02:27, Uri Blass wrote: >On April 30, 2001 at 18:48:55, Robert Hyatt wrote: > >>On April 30, 2001 at 18:28:56, John Merlino wrote: >> >>>On April 30, 2001 at 15:43:56, Robert Hyatt wrote: >>> >>>>On April 30, 2001 at 13:20:33, stuart taylor wrote: >>>> >>>>>I feel It's just a bit tooo inconsistent now. First 5 straight wins for DJ, then >>>>>nothing like it again, and even 3 straight wins for DF. >>>>>There must be something wrong somewhere! >>>>>People can say calmly "that's ok! nothing to worry about! It's quite normal!" >>>>>OR perhaps, chess is really a game of luck!!! >>>>>OK, if its proven beyond doubt that all this was the way things go, without any >>>>>imballance or bug because of things like learning, autoplayer etc. that's fine. >>>>>But I would like to see it investigated fully. >>>>> Unless it was that the first 5 games were openings which were advantageous to >>>>>DJ, but having "learned" them, DF was now well protected from these Junior >>>>>tricks. Could that be the case? >>>>>S.Taylor >>>> >>>> >>>>Pick any two programs. Play 50 games. You will find at _least_ one place >>>>where each won 5 games in a row. There is absolutely nothing unusual about >>>>this. >>> >>>Of course, you mean two theoretically equal programs (and do you also assume >>>that there are no draws?). And, just to test this, I just looked at the results >>>of a few 100-game matches, and there was only one instance of a program winning >>>four in a row. There were no five win sequences. There were MANY sequences in >>>which White won four in a row, though. >> >>I was not ignoring draws... and even with two programs that are 100 points >>different is skill, I have seen the weaker win 5 in a row in a reasonable >>match... > >It is not common Junior was 3 times black in the first 5 games. > >If you assume 30% chances for black to win and 40% chances for white to win >you get probability of less than 0.5% for Junior to win the first 5 games. > >The probability that one of the program wins games 1-5 is >0.4^2*0.3^3+0.4^3*0.3^2=0.00432+0.00576=0.0108 >The probability that one of the programs win games 2-6 is also 0.0108 >... >The probability that one of the programs win games 46-50 is also 0.0108 > >Conclusion: >The probability of the union of these 46 events is less than 0.0108*46 >It means that the probability to get 5 wins in a row for one side if you play 50 >games is less than 50%. I think the model is probably flawed a bit. I have seen some really strange sequences with chess programs. Especailly learning throws a monkey wrench into the model, since the events are not *really* independent. I wonder if the machines were swapped again after each set of 5 games. One machine being peculiarly better would explain the outcome to some degree if it was switched at game 5 and stayed there.
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.