As the title says, I’d be really curious to compare historical players to current ones using AI, much like has been done with chessmetrics. I found this link, but it doesn’t give the whole story. https://dl.fbaipublicfiles.com/elfopengo/analysis/www/index.html
Does anyone know a way I could analyse the accuracy of moves of different players, en masse, to test their points lost per move, and frequency of playing the AI top move and top 5 AI moves?
Probably Shin Jinseo.
That will require a lot of computing power, if you don’t get the data by someone who already analyzed historical games en masse.
I interpreted the OP as asking about pre-AG players, though looking back, maybe they did mean all
In some not very rigorous analysis I believe Dosaku got good marks from AIs, Michael Redmond did some videos on that.
But it was done at the time when KataGo was just started and LZ was still the main analysis AIs. Overall, it has a lot more variations than most thought. Unlike chess, there are some situations where many top moves are equally valid and the same (like in semeai, when you start to remove outside liberties, and most of the time all of the moves removing liberties would be the same). And Go AIs have evolved and changed a lot over the years, not even stabilized yet, and different AIs would give different top candidates, so if you just take the top candidate or top 5, they would give different answers (and different playouts would give different answers)
Overall, the historically famous players have roughly similar scores at the time of the analysis (since players from the past would have records spanning their whole lives, while modern players and some of the top players are mostly still in their prime. Comparing just a period of their lives with some ancient players’ whole careers isn’t that fair. Also, there is the problem of different rules, ancient players especially ancient Chinese players, would need to consider “group tax” and no komi hence when the analysis done, there haven’t be specialized trained AIs to deal with rule differences. Later on there are studies shown that the opening would change if applying ancient Chinese rules (one of the obvious thing is the opening 3-3 is a bad idea using ancient Chinese rules, since it creates a separate group and give yourself extra group tax, and the connect disconnect makes a lot of the early joseki branches to different direction, since cut off the opponent would worth points)
