Does anyone know what the rating of Stockfish is at various search depths (from 1 to 20 ply). I'm interesting in the ratings on a "human scale", not compared to other engines.
I can't give a specific answer for Stockfish but in a quick Google search this older article came up. For the older engines mentioned they mention 150-200 rating points per ply. Not sure how accurate that really is but it is a very interesting question.
Thanks. Yes, I've read that article too. But I'm guessing it's different today.
Another question I find interesting, related to this is: would it be possible to write a chess engine with a search depth of only 6 ply, but still being able to play at master level. I've read that most of the time even GM don't look move than 3 moves (6 ply) ahead. Still, they play at a rather high level. I expect this to be thanks to their positional understanding. Maybe an engine could be written to "think" in this "more human kind of way"?
It would probably be pretty easy to do a repeat of some of the types of experiments listed in the article. Using something like this (depending on how accurate that table is) and maybe Arena on a decent machine. Just would take a bit of time and a little number crunching :D
I don't know on the second question. I'm not very knowledgeable about engines (haven't done a lot of research). A quick search turns up that Rebel claims to have a "deep positional understanding". Maybe some others have a similar set of programming and could play strongly at a low ply. Don't know for sure.
It seems that a hard ply depth limitation would be a handicap. While GMs may only look ahead 6 ply in most positions, there are some where the look farther ahead, and it is possible that there are some positions where at 6 ply the move is great but at 8 it is bad.
Interesting table. Didn't know these numbers. Thanks.
It would take a lot of time. We are talking about thousands of games, if we want our results to be statistically significant.
I'm sure a search depth limitation would be a handicap, but it would be interesting to see how strong an engine could become, with a search depth limitation of e.g. 6 ply. Then, when it doesn't get any stronger, the search depth could be increased.