| |
BGonline.org Forums
More than a MET
Posted By: Fabrice Liardet In Response To: More than a MET (Bob Koca)
Date: Tuesday, 23 June 2009, at 7:31 p.m.
Your example shouldn't be a problem. Suppose we know our distribution and that it tells that a single win should be worth 0.6 (I don't know whether that is realistic). Then a 2-points win should be worth 1.2. Suppose your rating is so much higher that your expected result by the Elo formula is 0.5 (or 75% on the 0-1 scale).
Then your Elo equity by not doubling is 19/36 * 0.1 - 17/36 * 1.1 , while after doubling it is 19/36 * 0.7 - 17/36 * 1.7. By doubling you win 2/36 * 0.6 (times whatever the coefficient) Elo points. Actually I was thinking about Elo performances rather than the Elo rating updating after each game, but I think it also works correctly in that case.
It still seems true that money play is probably the one where the formulas will work worst. The limits I see are that :
- The result distribution from a rollout doesn't tell much about the result distribution of a session where at least one of the players plays very suboptimally (and the probability of an astronomic cube is much higher).
- The result of a very unbalanced money session can easily exceed one normalized point per game, in which case the Elo formula has exceeded its boundary. This can also happen in a match play game, e.g. if the GG player wins a gammon most of the time.
Between strong players and not too far from the end of the match, I would expect that method to give pretty accurate results though.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.