“Fidelity” Website FAQ

Some technical information to supplement the Tue. 3/20/2012 New York Times article, in particular the top-ten list. See also General Remarks, including some responses to queries and items and concerns on other sites, and my Parable of the Golfers.

1. The list shows only the “Opens” category. I have analyzed many categories and almost all “major” matches, round-robins, Olympiads and other team events, Ambers and other high-level Rapids, engine tournaments, even the whole PAL/CSS series of Freestyle competitions. This totals over 28,000 human OTB performances, over 30,000 adding correspondence, engines, and centaurs.

2. All of this data is in Single-PV mode, and acts only as a scientific control for the Multi-PV data on which all statistical conclusions are based. The Single-PV data is objective and reproducible by running Rybka 3 on one core thread in the Arena GUI (except that the mentioned stalling problem occasionally begs a manual re-start).

3. “Moves analyzed” excludes turns 1–8, moves in repeating sequences, and turns where Rybka 3 to depth 13 already judged the side to move to be ahead more than 3.00 pawns at the opponent’s previous turn. The 120-move significance cutoff figures 4 games with 30 eligible moves per game; I call this the “Larsen Line” because Bent Larsen played exactly 120 eligible moves in the 1970 USSR-ROW match, agreeing on 84/120 = 70%, which is 13th on the master list. (That is to say, despite the famous 19-move loss to Spassky, he held down first board really well, with two wins and a draw in his other three games.)

4. The 71.6% topping the Opens list is 3rd all-time for human OTB, ahead of every performance by anyone not named Shirov or Kramnik. Next come Aronian, Kasparov, Ivkov in 1980, Kramnik twice including Debrecen 1992, and Fischer 70.3% in his 6-0 match win over Larsen. The Paris performance uses the reconstruction of the Feller-Lagarde game here; using the official PGN or throwing out the game yields a slightly higher figure.

5. For most Opens and recent Olympiads I have made a selection including all games by the leaders and their closest pursures, and other notable performances e.g. where the performance rating is +150 or more. Occasionally a non-selected player has enough games againt those selected to make the 120-move cutoff. The lists include such partial performances—not to mention cases like this year’s Moscow Open where only games on top boards are (so far) available. Istvan Csom at the 1989 GMA Baleares Swiss and Jure Skoberne at the 2010 EIC are that type—I may eliminate or complete them later, but right now I am interested in the fact that they have some negative selection bias in contrast to the positive bias on the selected players.

6. Here is my list of all World Championship performances male and female, as posted first here in Susan Polgar’s blog. As with the Opens list in the article, this is just the Single-PV result, not the Multi-PV work used in the full model.

The beautiful fact here is that the top two are women! Since #2 is Susan in 1996, and #3 is Vladimir Kramnik in the rapid playoff in Elista 2006, this is the opposite of any insinuation against #1 Hou Yifan—whose opponent was no slouch either, coming in tied with Bobby Fischer in 1972, and also tied with…another woman, from 1978. Hou Yifan also is tops in the average-error category, and my full model gives her an IPR in the high 2900’s for that match (with Humpy Koneru a creditable 2680), alongside Kramnik in 2000 and Bobby over Tigran Petrosian in 1971.

7. I intend to publish the Single-PV part of my site, which is basically non-confidential, but it will take work to create a user-friendly front end, and there are some factors to iron out. The work itself is described in my published papers here, specifically the latter three.

More to come…

http://www.cse.buffalo.edu/~regan/chess/fidelity/FAQ.html

Chess Daily News from Susan Polgar
Tags: ,