ugh so many blitz and rapid tournaments in the chessbase database that the results are worthless. and doing more complex queries is taking ridiculously long. there has to be a better way

For some reason the program crashes whenever i try to simulate it any number of times. Managed to get it working though.

specifics of program:
-gave white a 50 point advantage
-bumped magnus to 2900
-bumped radjabov to 2810
-used the formulas from above to calculate how often win/draw/loss likely and randomized result
-did not adjust the ratings after each game and just left them static as they were at the start of the tournament

player 0 = aronian, 1 = gelfand, 2=ivanchuk, 3=svidler, 4=kramnik, 5=grischuk, 6=radjabov, 7=carlsen

Player 0 : 7.0
Player 1 : 5.0
Player 2 : 5.0
Player 3 : 8.0
Player 4 : 9.0
Player 5 : 7.5
Player 6 : 7.5
Player 7 : 7.0

Player 0 : 8.0
Player 1 : 6.0
Player 2 : 6.0
Player 3 : 7.0
Player 4 : 5.0
Player 5 : 6.5
Player 6 : 8.5
Player 7 : 9.0

Player 0 : 7.0
Player 1 : 5.5
Player 2 : 7.0
Player 3 : 6.5
Player 4 : 7.0
Player 5 : 7.5
Player 6 : 7.0
Player 7 : 8.5

Player 0 : 6.5
Player 1 : 5.5
Player 2 : 6.5
Player 3 : 6.0
Player 4 : 5.0
Player 5 : 7.5
Player 6 : 9.0
Player 7 : 10.0

Player 0 : 8.5
Player 1 : 6.5
Player 2 : 6.5
Player 3 : 4.5
Player 4 : 8.5
Player 5 : 7.5
Player 6 : 5.5
Player 7 : 8.5

Player 0 : 7.0
Player 1 : 6.5
Player 2 : 7.0
Player 3 : 5.0
Player 4 : 6.5
Player 5 : 6.5
Player 6 : 7.5
Player 7 : 10.0

Player 0 : 7.5
Player 1 : 5.0
Player 2 : 8.0
Player 3 : 5.0
Player 4 : 6.5
Player 5 : 6.0
Player 6 : 8.5
Player 7 : 9.5

Player 0 : 8.5
Player 1 : 6.5
Player 2 : 6.5
Player 3 : 6.0
Player 4 : 7.0
Player 5 : 4.5
Player 6 : 8.0
Player 7 : 9.0


after 8 trials carlsen wins 6, kramnik wins 1 and theres 1 three way tie between carlsen/aronian/kramnik. and just hitting run another 25 times, carlsen won another 20 times with the other wins going to aronian/kramnik/radjabov. nobody else ever wins. clicked through another 50-100 times just to see what the highest score he would ever get is and it ended up 11.0/14 a few times, probably 4 or 5 out of 100.

seems like carlsen is going to win 75-80% from my formula/program. if i change the ratings back to 2872 for magnus and 2793 for radjabov, magnus won outright 30 times, tied for a win 10 times, and lost to someone else 14 times. assuming he wins 70% of the tiebreakers, this would make him victorious 37/54 or 68.5%

either way i think -150 has value from my simulations. also seems like the only other person to have value on the given lines is radjabov as the guys under 2790 NEVER win and kramnik/aronian dont win significantly more than radjabov

the biggest problem with my program is that i think carlsen as black is an underdog to everyone and giving white 50 points doesnt reflect that. my guess is that expectation wise as white he expects to win 4 games, draw 3 games, lose 0 games and with black he expects to win .6 games, draw 5.7 games, lose .7 games. i think hes a slight underdog vs the field with black, but giving white 50 points means hes a favorite since he is rated at least 62 points more than everyone.

if i adjust the program to giving white 100 points advantage with carlsen at 2900 and radjabov at 2810 then:
magnus wins 47, ties for win 10, loses 12. assuming he wins 70% of tiebreakers, he wins 54/69 for 78.2%

every little adjustment i make just puts magnus smack in the 68-80% winrate.


making a guess based on the numbers i keep getting and some slight mental adjustments i would say magnus wins 70%, kramnik/aronian/radjabov win 7.5% each and the others are virtually drawing dead

Just looking at your calculation for the draw rate. Are you assuming that in a situation where one player is a 65% favourite, the favourite never loses, only wins or draws? If so, I think this would skew the results quite strongly in favour of the higher rated players.

I think a better way to do the calculation would be to assume points scored by the underdog to be distributed, say 50/50 (or some other proportion) between draws/wins.

So for a 65% favourite, the underdog scores 35%- half of the 35% would be from wins, so the dog wins 17.5%, draws 35% and the favourite wins the rest.

I have no idea what the right ratio would be, but you have to allow the possibility of big upset wins in your calculation.

Pairings are now up on the tournament website.

http://www.worldchess.com/candidates/index.html

This is just me complaining and being whiny, but the website design is pretty awful.

delete message plz

The web design is not bad if they paid $500 for the site. It wouldn't surprise me if they paid 20 times that though.

changed the program slightly

-gave carlsen 2880
-gave radjabov 2810
-all others their given ratings
-gave white 100 point boost
-gave underdog 2% to win from p=.63 to p=.78. couldnt up the win percentage more than that because then you really have to down the draw percentage quite a bit. but i think this gives a more accurate depiction of underdogs randomly winning sometimes.

so for instance, if gelfand (white) is playing carlsen (black), carlsen has a 40 point rating advantage after the boost to white which corresponds to .56 which gives carlsen 21% to win, 70% to draw, gelfand 9% to win. and for something like kramnik (white) vs grischuk (black), kramnik has a 146 rating advantage which corresponds to p=.7 which means kramnik wins 42%, they draw 56% and grischuk pulls off a miracle 2%.

given these new numbers carlsen won 50, tied for a win 10, and someone else won 13. again assuming he wins 70% of tiebreakers, that gives him 57/73 which is 78%. every time i make a change nothing changes at all

I built a simulator in Excel today at work. Finally have a chance to post results after a long evening. First of all the criteria: I gave white a flat +35 rating points, and set the draw rate at a universal 55%, based on things I read on Wikipedia.

At this time I do NOT alter result odds for any kind of momentum factor, and I do not use player-specific numbers for the draw rate or white advantage. I would like to add these features, but that's for later. For now, I simulated the tournament 5000 times. There was a clear outright winner 4168 times, 684 two-way ties, 129 three-way ties, 18 four-way ties, and one five-way(!) tie - how awesome would that be? In the case of ties, I figured all bets are off for tie-breaks, and gave each player tied for first equal win equity.

Here are my results:

I'd much prefer to have the easier games at the end if you're a top tier. Don't forget that your opponents will also be drained and there's psychology kicking in as well. The bottom half in an event this stacked is probably going to end up getting shredded. Going +0 =5 -3 is not going to leave most people in a particularly positive mindset when facing one of the best players in the event. And on top of that they are no longer even competing for a chance at being the candidate, they're just left going through the motions to finish the tourney.

Actually this is even a testable hypothesis. You could simply look at the performance vs expectation of top seeds in later rounds of double round robins when facing opponents who are already out of contention. 'Simply' being a bit of a poor choice of adjectives though as this would rather be a pain in the ass to collect and organize the data for.

I got Carlsen as a lock to win this ****.

Any online betting available for this?

I got the odds off this site.

https://www.unibet.com/betting

I assume they offer online wagering.

http://www.marathonbet.com/en/betting/Chess/

Also Betfair have priced it up where Carlsen was interestingly matched at a high of 2.4

Doesn't seem to factor in draws, though, which isn't good enough in my book. It says Carlsen is 43% to win this event, but sets him at 22% to score 9.0, and 19% each to score 8.0 or 10.0, with no chance of 9.5 (or any other half point increment). It appears to just calculate an expected score, an assume a normal distribution around that score. Without draws. NOT GOOD ENOUGH!!! OMNOMNOM PROGRAMMING!!!

I like those portrait articles. I will use them for each player's draw rate and white/black score differential. As soon as the 8th article goes up, I'll re-run the sim with those stats in. I got the program re-tuned to accommodate it today. I also updated the formula to allow decimal values for player's ELO ratings, and replaced everyone's March Rating List rating with their Live Rating instead, which cost Kramnik a good bit of win equity, and made Carlsen even a bigger favorite, but the differences weren't significant enough to go into detail on, until I have play-style stats on all eight players.

I've decided I don't want to add any kind of "momentum" concept, I will work on the theory that what appears to be "bad tournaments", from players who are "out of form", it's actually just human application of a false narrative to results that below expectation due to random chance.

That said, I have considered factoring in something similar. I want to add a function for "playing for a win", for players who are trailing the leader, but only by a small margin, in late rounds, and for "playing safe" for the leader in that same situation. In either case, it would be modeled as a reduction in ELO rating, in exchange for either a reduced or increased draw rate, as appropriate. I'm not sure if I will actually get that built in, or for that matter whether I should, and would be interested in input. Good idea? Or overthinking things?

Nice Magnus picture in Guardian preview.



http://www.guardian.co.uk/sport/2013...en-world-title

Round one starts on Friday, just a few more days now. Can't wait.

Accd to that article Carlsen v Aronian on first day? I'll be renewing that ICC subscription for the commentary, I think

So those chessbase player profiles are coming out pretty slowly. Are they going to get the last four posted at a quicker pace, so that all eight are out before the tourney starts? If not, my plan of using those to calibrate my simulator won't work so well.

(Although one of the cool things about the simulator is that I will be able to put in actual results, and do monte carlo simulations of only the unplayed rounds, and offer updated odds of each player winning as the tournament progresses. So even if my data isn't sound until after the fourth round, I can still do something with it.)

yes, i believe so at this link

http://london2013.fide.com/en/compon...137-live-video

The simulator sounds awesome. Rd 1 is on Friday and looks like Chessbase is on pace to publish the Magnus profile on Thursday

I don't see how you calculate that pace, with four of eight profiles published so far, on the 4th, 6th, 8th, and 11th of March. Even if they put up #5 today, moving to a one/day pace for the last four, that still doesn't get Magnus up until Friday after the first round has probably already started.

As for the simulator, it's just a Macro-Enabled Excel file with a few small VBA subroutines in it for things like "simulate a game", or "simulate a round" or "tabulate results". Right now it's hard coded for this particular tourney, but it would be pretty simple to re-work it for any tournament with a pre-set schedule (trying to predict the results of a Swiss event would require a lot of new coding).

I would be happy to share the file with anyone who wants to explore it. If you know VBA it shouldn't be hard to understand (and if you're actually a "real" programmer, you'll no doubt be able to find all sorts of spots where I did things the "hard" way, hehe). The only challenge is the time delay, I can't check 2+2 at work(blocked as a "gambling" website), and I can't touch the macro at home (don't have the full version of Excel at home). So shoot me a PM if you're interested, and this evening I'll tabulate email addresses, then tomorrow at work I'll send out copies.

Ouch, I thought there were only three more to go. *waves fist at chessbase*

ha! http://en.chessbase.com/Home/TabId/2...le-120313.aspx has a link to a spreadsheet with all the matchups and results

I saw that spreadsheet, and was initially excited, but I'm reluctant to include blitz/rapid results. It does list what event each game was played in, but manually splitting the fast time controls out is a project I am not eager to embark upon. Also, it is not pre-sorted by color, like I'm looking for. Again, it would be possible to extract the data in the form I need, but not quickly and easily.