2

I am thinking on a way, most ideally on a software tool, what could get one of more games of a player, analise them, and gives an estimate on his play strength (as Elo/Glicko rating).

Is it possible? Does such a tool exist?

Gray Sheep
  • 135
  • 4
  • 1 - by a computer: maybe average centipawn loss (assuming you adjust for time control) ? 2 - by a human: do you know guess the elo by gotham chess? – BCLC Sep 03 '21 at 00:27

1 Answers1

3

Yes it's possible. Sort of. It's one way people have compared the top players throughout history. The basic idea is that you feed the engine the position and the moves played by the players, and see how probable it is for them to play the computer's best move.

I don't know any ready-to-use program for this kind of analysis, however.

Allure
  • 25,298
  • 1
  • 65
  • 143
  • 2
    I always encourage to read the primary literature. Newest computer and method: https://content.iospress.com/articles/icga-journal/icg0012 (article seems to be free) – Hauke Reddmann Aug 28 '21 at 08:30
  • Ok, but if the game is already quite trivial, it causes a shift into the positive direction. For example, the chess.com custom is that nearly everybody plays until mate and never gives up, because they are a...s. It seems, the readiness to waste the time of the opponent causes a comparative advantage there. Imho this calculation should somehow consider also the hardness to find the current best step. – Gray Sheep Aug 28 '21 at 19:22
  • 2
    @GraySheep that's why one of the sentences in that article is "A 2008 analysis, using Rybka 3, showed that Capablanca had the smallest average error factor (i.e. the most accurate play); but *after adjusting for factors such as the complexity of positions*, the best player came out as Fischer, followed by Capablanca, Karpov and Kramnik." (emphasis mine) – Allure Aug 29 '21 at 00:51
  • @GraySheep "_nearly everybody plays until mate and never gives up, because they are *****_" Even titled players encourage amateurs to play to mate (GM Ben Finegold, IM Levy Rozman, and others). I don't think your comment is evidence-based, and it would really depend on more details (rating, etc., as suggested by these titled proponents). – SecretAgentMan Aug 30 '21 at 12:51
  • It is very hard to evaluate how well the algorithm performs when comparing World Champions, almost any result looks reasonable. Has this approach been tested on something that is easier to evaluate, for example a confusion matrix of Category C, B, A, Expert, Master and GrandMaster? – Akavall Aug 30 '21 at 18:22