Please turn on javascript in your browser to play chess.
Only Chess Forum

Only Chess Forum

  1. 06 Jan '17 19:56
    I've made something which I think could help explain some of the differences between players: a visualization of different statistics. Any players you would like to see me do this for? Anything I should include or exclude in the future? Leave it below.

    Blog 348

    Thanks,
    Hikaru
  2. 06 Jan '17 21:22
    Hi, an idea to solve the non-transitive property as you mention. If you can define a reference number for each category, you only need to compare to those. It will allow you to make the radar plots, as you only compare to the "reference player" and not a specific real player. The reference numbers can be based on the average over a large, representative group of players. They will end up as weighting factors in your calculations, so they don't even have to be accurate or up to date.

    I like the overall idea of being able to classify a player based on objective criteria. Commentators always throw around with playstyles et cetera, but I am left to just believe them, because they don't come up with objective statistics.

    The choice of criteria may be a bit bothersome. I didn't understand all of them (either the term or the definition) and they may be subject to criticism. I hope you will get input from others about these definitions/criteria.
  3. Standard member patrickrutgers
    Pale Yellow Star!
    07 Jan '17 01:23
    Nice Fresh Prince of Bel-Air reference!
  4. 08 Jan '17 19:57
    Hey @tvochess, thanks for the feedback. For the radar charts, I think the reference player might be a good idea. The other way I was thinking of was to compare all the players to each other. Say you have a 20-person group. Then you could sort players by their percentile in the group. The only concern is whether this would be effective with such a small player pool.

    The commentating problem is something similar to what I'm trying to address. It's much easier to say that Giri draws a lot than say that he has the lowest winning percentage as White. (Or, for that matter, get Giri's draw rate, which is fairly easy. It's 0.419, much higher than someone like Carlsen's 0.261) It's also much less punchy, and harder to back up under actual analysis. (By the way, the White in % thing is true: https://www.reddit.com/r/chess/comments/5m5vt3/some_stats_on_the_top_100_players_in_2016/) These things are quantifiable, and I'm trying to make them easier to understand.

    Which is most of why I'm also concerned about the criteria: I want them to be easier to understand, and I want them to be actually reflect something. So some were drawn from sources like chess-db.com, which has some statistics (which I'm a little uncertain about, but seem credible,) and some were from chessgames.com, another opening database.

    For example, the opening popularity: for the four most-played openings for White and Black, I went and clicked through the most-played moves from both sides for five moves. Then I determined what % of the time when Carlsen was faced with these positions that he encounters often, he plays the most popular move. I also hope I will get more input, and I hope that clarified a few things.