Forums

game review player ratings

Sort:
tdstrnbrg
You know how at the bottom of that tally of moves by brilliant/great/best/etc, it rates both players? I've found that those ratings are influenced by the players' actual ratings as they appear in the PGN file. I took the PGN of a game by two grandmasters and, editing their ratings, reduced them by 2000 points. Now ran game revirw again, and their performances got rated about 1000 points lower. Not really a big deal but sorta interesting.
Conxv

hmm its a very intresting discovery. that explains why my current games are rated higher than my old games even when my old games are better blogs

Duckfest

That's something I've noticed as well. Just like you I've experimented with adjusting the PGN to see what would change. These were my main findings:

  • Estimated player ratings in the game review are dependent of player ratings, of both players.
    • Giving my opponent +1000 rating gives me a bonus of 450 rating and my opponent a bonus of 500. Total rating increase of 950 
    • Giving myself a +1000 rating gives + 150 rating for me and 400 for my opponent. Total +550.
    • Giving both players +1000 rating gives 500 rating bonus and 850 to my opponent. Total +1350 / 2000.
    • These numbers are dependent on the initial ratings, your milage may vary. But higher rating will not directly impact the estimate in the same amount. Which makes sense, if you played like a 1400, the game review won't suddenly consider you a 2400 when you increase the Elo rating of the PGN.
  • Ratings are not dependent on the accuracy of the other player.
    • Playing 100% accurate in a game under easy circumstances, like blunders, gets the same rating performance as accurate play against an opponent that plays very accurately.

It's interesting that your rating estimate will go up if your opponent is higher rated, for the exact same game. However, if you opponent plays a perfect game, it won't impact the estimated rating.

Squirrel

Hi, thank you for your reports. The main reason (I believe) we try to adjust estimated ratings according to players' ratings is because it would be rather confusing to see a 12-move game between two players rated 500, show '2500' rating for one of the players, just because they played a gambit. It would spark more 'This guy is cheating, your engine says he played like a GM' comments, which do not help grin.png The current estimated ratings help visualize the level of the game compared to the average assumed strength of both players. In any case, I think this is good feedback to keep in mind as we improve Game Review, if you have any ideas how we could improve it more, let me know!

tdstrnbrg

@Squirrel: I think it's fine that you do it like this. You might call it a Bayesian approach.