< Earlier Kibitzing · PAGE 31 OF 31 ·
|Aug-08-12|| ||achieve: <alexM.> I went on to read Sonas' I must say very transparant and well written explication on his "weighted and padded" (as you said with the 4-7 fake games added) TPR calculations, and I find he is doing a very professional job with it. He explains the apparent nonsensicality by informing us that he has done extensive adjustments in order for the number to be best at predicting future scores, und so weiter, but I agree no specific data or formulas or access to his data....|
Thank you for the quick and elaborate response, and I will dive into it a bit later, and this already is extremely helpful. :)
|Aug-08-12|| ||achieve: Oh, one more thing springs to mind... Sonas in the past advocated to improve the formula by setting the K-factor to <24>, as opposed to 10, 15 or 30...|
If you have time I would like to know how you "handle" the K-factor (as on the Fide site calculator) in various circumstances, depending absolute and relative Elo/differences.
|Aug-08-12|| ||alexmagnus: I don't have an established opinion on the "correct" K-factor yet. There are rating systems with a "movable" analog of K-factor - like Glicko. Which is interesting, as a K-factor depending on activity solves the "problem" of inactive players starting with their ratings again even after not having played for 10 years (like it was in Kamsky's case). But the problem with a changing K-factor is that the games stop being zero-sum, and so we know nothing about the stability of such a system. Elo is at least mid-term stable (about its long-term stability the debate will probably never end).|
As for a constant K-factor, I think the present system is OK. It takes into account that the players on the higher end of the ladder are mostly professionals (and therefore more active), giving them a lower K-factor. Making it bigger for everybody would only make sense if the ratings get updated daily (and not monthly as now).
|Aug-10-12|| ||achieve: <But the problem with a changing K-factor is that the games stop being zero-sum, and so we know nothing about the stability of such a system.>
A variable K-factor indeed seems ludicrous, exactly because of what you say... Sonas advocated (in a Chessbase article) the K-f to be 24, I think only to be applied at master level high frequency chess. But I'm not sure, as his article was rather long, somewhere from 2002/3 I guess, so at least dated, but he might have suggested the K change to be applied uniformly, establishing a Golden Mean with more accuracy for the top echelon.|
I'd have to read it back again, but atm my time is limited.
Thanks sofar, <alex>, you've been very helpful, and I may visit soon again, as I still have some left over things I'd like to understand better.
Oh - this one, <Elo is at least mid-term stable (about its long-term stability the debate will probably never end).> -- I have an interesting graph for you:
Rating developments comparing the #5, 10, 20, 50 and 100 of the list from 1975 to 2009: "stable" from '75 til '90 -- after which there appears a steady, ("inflatory") climb upwards, which over the last few years seems to have halted.
|Aug-10-12|| ||alexmagnus: I've seen that graphs, but there are some problems with Sonas' arguments us that he changed his definitions of inflation on numerous occasion (shortly after that graph came out he brought another article, at which he himself was surprised how good the Elo curve fits the actual results (and admitting his own "85-point-formula" was based on looking at elite games only. In that article he said he gained some new insights into the rating inflation problem but didn't say what these insights were).|
As you probably know, I myself think there is no inflation in terms of the rating/skills ratio. The upward trend (and its slowing in the 21st century) I explain by two phenomena - the "dying generation" and the "dominator effect" (both terms are my own). A third factor, computers, is disputable.
"Dying generation" refers to the situation in the mid-70s/early 80s, when there were more top players aged 40+ than those below 40. This circumstance, on the one hand, opened a possibility for Karpov's dominance, and on the other hand, for the "room to above". As most top players during that time were past their peak, it's not surprising they (and their ratings) were overtaken when a younger generation appeared.
As for the "dominator effect" - in any sports (with absolute measure of skills) one can notice an interesting phenomenon: when a dominant player appears, other players start to catch up with him and improve their results. The dominator may "run away" - this cat-and-mouse game lasts until the dominator "runs out of power". Isn't it strange that the upward trend in the top-100 ratings began in 1986-1987, when Kasparov started dominating (as opposed to "just" being #1)? And that slowing of that trend began with the fall of Kasparov's rating?
The third factor, computers, may have contributed in multiple ways, but all are disputable. It's indisputable that computers and the Internet raised the <average> level of chess worldwide. The question is - did it also affect the elite level? If yes, it's the third factor.
The other computer-based factor is the change of the generational model. Players can reach a respectable level at a younger age and retire at an older age - because they can keep their skills with a computer training and preparation, as efficiently as never in the pre-computer times. I can even imagine we will witness another "dying generation phenomenon" somewhere around 2025.
As for "if all players improved the ratings wouldn't change" - correct, but only if the improvement was simultaneous. But at least in case of the "dominator effect" it wasn't - it normally works so that first only the direct concurrents of the dominator improve, then the concurrents of the concurrents etc..
|Aug-11-12|| ||frogbert: < Sonas advocated (in a Chessbase article) the K-f to be 24>|
achieve, interestingly, when fide was discussing doubling the k recently, sonas was part of the expert group that discussed this possibility. and eventually sonas ended up deciding against his own original suggestion (increasing the k factor). if i remember correctly, simulations showed that such a change would lead to notable (systemic) inflation compared to the values in use today - which probably was the main reason this change never happened.
|Aug-12-12|| ||alexmagnus: <frogbert> Did Sonas <ever> cared about the systemic inflation? I don't remember him defining inflation in a way similar to your understanding of systemic inflation. But he changed his definitions so often that I may well have missed one or two of them :)|
|Aug-12-12|| ||frogbert: no, i don't think he did - but whether the "expert group" realized or not, it was exactly *systemic* inflation their simulations with higher k's demonstrated.|
i assume you understand why, but to me it was kind of amusing that the fide experts (including sonas) kept coming up with creative theories to explain inflation even *after* running a simulation that pinpointed the slight imbalance in the system that fuels a very moderate *systemic* inflation.
|Aug-12-12|| ||achieve: This is over my head by some distance, guys, but these "changing definitions" by Sonas, they surely are documented somewhere? Sonas took a break from his chess related research and returned in 2009? Is there as per frogbert's account a link to this recent meeting of the expert group that included that Sonas vagabond?|
<it was kind of amusing that the fide experts (including sonas) kept coming up with creative theories to explain inflation even *after* running a simulation[...]>
A website? A link? Microfilm?
Not kidding, I'm interested to read up on some of that.
|Aug-12-12|| ||achieve: Aha, my seismograph is ticking here:
Doesn't Sonas look an awful lot like actor John Goodman?
Have to peruse this later, and there are some useful links to related 'expert' articles from 2011, I noticed. But basically you two are saying that their work is flawed "to a degree"?
|Aug-12-12|| ||alexmagnus: <but these "changing definitions" by Sonas, they surely are documented somewhere?>|
Three definitions of inflation used by Sonas in different times:
1)Comparing average rating of <same> players ranked 3 to 20 on both lists. This is how Chessmetrics lists were "calibrated".
2) Comparing rating of number X. This is the definition used un the article you mentioned (where he shows the upward trend)
3) Unknown definition used for evaluating Carlsen's performance at Nanjing 2009 (one of the ugliest articles ever written by Sonas, trying to create a senseless Elo-Chessmetrics hybrid). He corrects there 29 points for inflation (between two different systems, a comparison useless in itself), saying that it would take too long to explain how he did it. It was neither of the methods mentioned above, at least I couldn't figure out which one (when asked about that article in personal correspondence, I got this as response: "The Carlsen article was
requested by Chessbase on short notice and so I did the best I could in the time I had."). Still unclear how he got to that 29 points though.
|Aug-12-12|| ||frogbert: achieve, the two "important" articles by sonas that i had in mind was the one with newsids 5608 and 5527. the former links to the latter. i'll get back to you later, got two kids to watch.|
|Aug-12-12|| ||achieve: Thank you both, and that newsID-link is a great idea, frog, very handy, quick, and accurate.|
Planning to be back later this evening.
|Aug-12-12|| ||frogbert: was just hard to copy the links on my phone, so i took a shortcut.|
|Aug-13-12|| ||achieve: I wasn't being ironic! I just clicked my link and replaced the ID number in my browser, et voila! I actually liked performing that "trick", being the relative computer hack that I am.|
|Aug-13-12|| ||frogbert: didn't think you were ironic, actually. just felt like explaining the reason why i was so "creative". :o)|
|Dec-12-12|| ||mehdimike2: hi alex,u r the most objective user in this site!:)
Enjoyed many of your comments.
From a 22 yeras old persian boy
|Dec-13-12|| ||alexmagnus: <drik> My weight looked quite dangerous to you, huh? I know, it shocks many. But I'm fine. And what's funny - it doesn't change whatever I do to my eating behavior or my body. 38-41, with 39 being the most common figure (at 170 cm height).|
|Dec-13-12|| ||drik: <alexmagnus: <drik> My weight looked quite dangerous to you, huh?>|
No - I'm Bangladeshi in orgin, so I've seen people of that weight pulling rickshaws & working on building sites. I know it is viable - but normally requires intense discipline or deprivation! I'm 165 cm tall & weigh 70 kg ... & now urgently feel the need to diet!
<And what's funny - it doesn't change whatever I do>
This is very interesting. Either your metabolic thermostat is permanently on maximum (is your resting heart rate above 70?), or you might have unusual genetics which might be worth profiling. Imagine the money you could make, if you could bottle it & sell it to the Americans! ..or the Germans!
As for changing it, you might consider the Sumo practice of eating chanko-nabe for lunch & sleeping just after. Hakuho used it to go from 60kg aged 15 to 150kg aged 26 ... though you might consider stopping, just where he started ;-)
|Dec-14-12|| ||frogbert: alexmagnus, carlsen is set to equal korchnoi's 8th place on your dominator list. maybe you need to start numbering the items in case of shared places?|
|Dec-14-12|| ||alexmagnus: <frogbert> Normally I order the shared places chronologically (as you can see in the case of the Tal/Ivanchuk/Anand trio). Not intending to say that the earlier dominance is "better", just to have some convention.|
|Dec-15-12|| ||frogbert: actually, my opinion would be that later dominance (by the same amount) would be "better", if anything. simply based on the presumed increase in strong players. anyway - hardly very important.|
|Dec-18-12|| ||whiteshark: <Alex> Hanna-Marie weiterhin im C-Kader. Was auch immer das bedeutet... http://www.chess-international.de/?...|
|Mar-03-13|| ||frogbert: carlsen inches forward wrt "dominance over #2", with 63 points in the upcoming april 1st list - because aronian will (still) be at 2809 then and carlsen at 2872. aronian's 2 bundesliga games will probably not be rated until may, and then we'll also have the candidates. kramnik, though, will be down to 2801 by april 1st...|
distance to #10 will be 106 due to caruana's climb, so next possibility to improve his record there for carlsen will be the may list. but honestly i won't care very much as long as he wins the candidates. :o)
|Mar-04-13|| ||alexmagnus: < but honestly i won't care very much as long as he wins the candidates. :o)>|
If he wins it in his usual "winning" way it (breaking own records) will happen anyway :).
< Earlier Kibitzing · PAGE 31 OF 31 ·
Bobby Fischer Tribute Shirt