Very nice article, good work on the data analysis and visualizations. It's great to watch these matches here. It makes some sense that they start out carefully. Perhaps we'll see some wins and losses as the match goes on.
Interesting read. Cool to see what games were the most accurate but otherwise mostly confirms what we already know (chess is getting more accurate etc), but using just a depth of 20 seems very low to me.
Very weird things happen at that depth (comp considers the King's Indian worse than it really is for example), and it seems from some previous posts that multiple games were input incorrectly and so it does call into the question the correctness of some of the results (if these posters are correct, though of course the overall conclusion that modern players are more accurate is of course correct as that has been confirmed with other studies).
Very weird things happen at that depth (comp considers the King's Indian worse than it really is for example), and it seems from some previous posts that multiple games were input incorrectly and so it does call into the question the correctness of some of the results (if these posters are correct, though of course the overall conclusion that modern players are more accurate is of course correct as that has been confirmed with other studies).
@dRr0x0rZZ said in #5:
> Geez they're gonna need to add some squares to the chessboard to liven things up...
Why not! For real, why not.
Chess -> Ciess (because i comes after h)? And introduce new minor piece, what could that be?
> Geez they're gonna need to add some squares to the chessboard to liven things up...
Why not! For real, why not.
Chess -> Ciess (because i comes after h)? And introduce new minor piece, what could that be?
I wouldn't attach too much value to ACPL as a measure of accuracy. Surely long prepared computer lines keep this average low, but long drawish endgames have the greatest effect. Often the evaluation remains very tightly around 0.00, and many different moves keep the balance. So a long endgame without blunders greatly improves the ACPL for both players. Furthermore the types of positions on the board also matter a lot: it is easier to be accurate (according to the engine) in forced lines than in quiet positions. Finally, if one player gets a significant advantage, then there are often multiple ways to win, and not playing the computer's first choice there will increase the ACPL.
I don't know how each of these points affects the ACPL scores in this dataset, but I just want to stress that ACPL is way too simple a statistic to assess the quality of play.
Still, this article shows a trend in the ACPL over time, which is interesting. But it is unclear to me what the conclusion should be.
I don't know how each of these points affects the ACPL scores in this dataset, but I just want to stress that ACPL is way too simple a statistic to assess the quality of play.
Still, this article shows a trend in the ACPL over time, which is interesting. But it is unclear to me what the conclusion should be.
A truly silly thread idea.
@RoyalHog said in #7:
> The most accurate games will be those that end as a draw right after the opening and are essentially home analysis...
Agree, also in WC they have teams working for them. Lets say - it's more fun to analyze games where the players aren't trying to pull off the soundest moves. Why not the candidates, there's controversies when players (who are not friends) face each other, no easy draws then. Ex., Carlsen's games in the candidates 2013 or Caruana's games. In 2016 candidates he drew a crucial winning Rook and Bishop vs. Rook vs. Svidler, then tried to defeat Karjakin and lost. Now such games are interesting for somebody's trying to win.
> The most accurate games will be those that end as a draw right after the opening and are essentially home analysis...
Agree, also in WC they have teams working for them. Lets say - it's more fun to analyze games where the players aren't trying to pull off the soundest moves. Why not the candidates, there's controversies when players (who are not friends) face each other, no easy draws then. Ex., Carlsen's games in the candidates 2013 or Caruana's games. In 2016 candidates he drew a crucial winning Rook and Bishop vs. Rook vs. Svidler, then tried to defeat Karjakin and lost. Now such games are interesting for somebody's trying to win.
Interesting read, Capablanca was well ahead of his time...
Not really to find most accurate player but as yet another comparison between WCC matches and get some insight from it.
Not really to find most accurate player but as yet another comparison between WCC matches and get some insight from it.
What about a different measurement, average centipawn loss / moves played?
Definitely definitely definitely me ;)
There is significant evidence to indicate that various games from the data have the wrong moves and should be reexamined to find the correct moves, which would likely lower the ACPL of those games. In addition, it would be interesting to compare the accuracy of the games from both championships during the time period with disputed titles. Finally, after the games are corrected, they should be reanalyzed with an extremely powerful computer.
This topic has been archived and can no longer be replied to.