Quote:
Originally Posted by BobJoeJim
I would be interested in running some sims. I would never consider doing so without including draws. Can anyone dig up stats on what percentage of serious games each player has drawn, over, say, the past two years? "Serious games" can mean whatever you think it should mean, but obviously not blitz, and probably nothing against anyone rated below 2600. White/black splits too, please, if someone consistently plays for the draw (and achieves it) as black, and then plays much more aggressively as white, I would want to account for that.
From color-specific draw rates for each player, and ELO ratings, I can put together a formula to estimate the odds of win/lose/draw for any pairing. From there it's just monte carlo time, although I may add a function to the sim that calculates an "effective" ELO rating for later rounds, based on some combination of actual rating and results so far, to model certain players being in good or bad form. If I put any sort of form/momentum factor in, I'll be sure to run the sim both with and without it, to satisfy both those who do believe in momentum, and also those who think each game is a truly discrete event.
cant get c++ to work on my computer. gonna give up. but heres the template i was going to use for my idea. hope this is understandable, might be a bit confusing
i was going to give everyone their given rating except carlsen who i was going to give 2900 since its still rising somewhat steadily and i assume it will asymptotically approach 2900 given his performance ratings over the last few years.
i was going to give the white player a 150 point rating boost each game and use this chart in article 1.48a to figure out the p given rating differences:
http://www.fide.com/component/handbo...4&view=article
p is just % of the expected win. i wrote my own formula for win/lose/draw from that.
for p = .5, i assumed 15% player 1 wins, 70% draw, 15% player 2 wins.
p = .51, 16% player 1 wins, 70% draw, 14% player 2 wins.
p = .52, 17% player 1 wins, 70% draw, 13% player 2 wins
...
p = .65, i assumed 30% player 1 wins, 70% draw, 0% player 2 wins.
just gave a linear progression for those p values.
for p = .66, 32% player 1 wins, 68% draw, 0% player 2 wins.
p=.67, 34% p1 wins, 66% draw
...
p=.86, 72% p1 wins, 28% draw
so for instance, carlsen vs gelfand where carlsen is white. i give carlsen a 150 point rating boost to 3050. the rating difference is now 310. look up on the chart that means p=.86 which means carlsen is 72% to win, 28% to draw. whereas if gelfand was white, he gets a 150 rating boost to 2890 for a rating difference of 10. thus carlsen's p is .51 which means hes 16% to win, 70% to draw, 14% to lose. also seems reasonable
so basically i was going to enter a lookup table for p based on rating difference, then figure out how often they are going to win/lose/draw based on that p, randomize it, adjust those players ratings after each game using a simple ELO adjustment formula and then run this for every matchup between every player probably 10k times and see how often everyone wins.
thoughts?