home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: rec.games.go
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!sdd.hp.com!sgiblab!cs.uoregon.edu!news.uoregon.edu!news.u.washington.edu!stein.u.washington.edu!adrian
- From: adrian@stein.u.washington.edu (Adrian Mariano)
- Subject: Re: winning ratio
- Message-ID: <1992Nov21.004103.14576@u.washington.edu>
- Sender: news@u.washington.edu (USENET News System)
- Organization: University of Washington
- References: <1992Nov17.023826.2810@bhprtc.scpd.oz.au> <1992Nov18.021006.14737@uxmail.ust.hk> <1992Nov18.060917.27210@u.washington.edu> <1992Nov20.082459.6464@uxmail.ust.hk>
- Date: Sat, 21 Nov 1992 00:41:03 GMT
- Lines: 92
-
- In <1992Nov20.082459.6464@uxmail.ust.hk> schmidt@uxmail.ust.hk (DR. ROY SCHMIDT) writes:
-
- >adrian@stein.u.washington.edu (Adrian Mariano) writes:
- >>schmidt@uxmail.ust.hk (DR. ROY SCHMIDT) writes:
- >>
- >>>>>
- >>>>>P = 1/2 * (2/3)^(2*d)
- >>>>>
- >>>>>where d is the absolute rank difference and P is the probability
- >>>>>that the weaker player will win.
- >>
- >>>There are several problems with such a "formula" approach.
- >>
- >>>1. If we assume no difference in ranks, then the formula yields a
- >>> probability of 0.333 of a given player winning. Since the game is
- >>> not three-handed, we would have to assign the remaining 1/3 to draws
- >>> Having played not a few games in my life, draws just don't seem to
- >>> come up often enough :-).
- >>
- >>No. 1/2 * (2/3)^0 = 1/2. This is the expected result.
- >>
-
- >Oops! There seems to be a line missing here! My post should have read
- >that the formula yields a .50 probability of a given person winning, but
- >if we are being even-handed, we should assign a probability of 0.333 of
- >a given player winning. After the word "Since" it is correct, and an
- >attempt to inject some humor into this thing. p = 1/2 is not the
- >expected result, because there must be an allowance for draws. We did
- >say *no handicaps* right? I assume this means no komi, as well.
-
- No handicaps would suggest a 5.5 komi to me. That's the standard komi
- for even games (at least, in some places). There are no draws to
- worry about.
-
- >>>4. The formula lacks an empirical basis. First, you have to find a
- >>
- >>The formula was obtained from study of over 2000 games played in
- >>tournaments. The study was done by Jos Vermaseren. This doesn't
- >>conform to the (omitted) guidlines you demand, but it is the best
- >>emperical basis you can reasonably expect. It shows (to some degree)
- >>what happens when real people with real ranks (however they were
- >>determined) play games.
- >>
-
- >Hmm. The study by Joe Vermaseren you refer to had a different formula.
-
- No it did not.
-
- >That formula also did not allow for draws. Joe also admitted that the
- >difference in ranks seemed to be affected by extended play. I found
- >his reasoning to be a little hard to follow (because there was not much
- >presented, and some problems with the English used). Also, the idea
- >that a linear relationship (his conclusion) could be represented by an
- >exponential function does make one wonder. PLUS, I couldn't see
-
- If you really want a way to get linear graphs out of exponential
- functions, take logarithms.
-
- What Jos said (in rON's quote) about this is
-
- >function of x-y. I have done this measurement with a database of about
- >2000 tournament games a few years ago. Actually if x weaker than y, the
- >function P(x-y) = ((1/3)^(y-x))/2 with x and y in units of half handicap
- >stones and beginning gives only half a stone more on the board.
- >When this function was used for the probability in a giant maximum likelyhood
- >fit for all these results the graph of measured strength versus generally
- >accepted strength was linear (with some scatter of course) from 5 dan to
- >15 kyu! I had not expected this in advance and it is something I still
- >do not understand.
-
- I don't think there's enough information to figure out what's going
- on here. I don't know what the parameter for the maximum liklihood
- was, or what the "measured strength" was. It would be nice to know
- what distribution of ranks were present in the data. (Probably no 15
- kyus played 5 dans.)
-
- >anything in Joe's post that said that all 2000 games were played
- >without handicap.
-
- This was obviously true, or the games would not have been suitable for
- his analysis.
-
- >Perhaps Joe could help by showing us a little more of his data and the
- >regression(?) analysis he did for each of the differences in ranks.
- >At least what he has could be a starting point, but I rather doubt it
- >could be the conclusive result which you seem to think it is.
-
- My previous post expressed too much confidence in the result. It
- would be good to see more detailed information about the data and
- analysis.
-
-
-