3

Do transposition databases exist the way tablebases do? Would an engine be able to take advantage of a pre-calculated TD? Is there a benefit to players when studying if they had access to pre-calculated positions?

What I am imagining is a sort of Wikipedia for chess positions. Rather than focusing on perfect endgames or history of play, the emphasis would be on the position itself rather than the moves leading up to it. Users would then “kibitz” the individual positions, with popular ones rising to the top based on visits or appearance in games.

Does something like this already exist? I’m aware this is similar to some of the features of “Let’s Check” from Chessbase. I’m thinking open, like tablebases are.

And on that note, are there any proprietary table bases?

  • something like this chessdb.cn/queryc_en? TCEC once played a game with stockfish vs stockfish, but one used this database as a book (similarily to how brainfish does). As far as I remember it was worse. – Sopel Dec 13 '19 at 9:35
  • Great suggestion. This is very close to what I am imagining. It's just lacking the social features. I also like that it can interface via UCI. – Justin Dec 13 '19 at 10:09
1

What you're talking about is basically opening books. If an engine has access to hard-set evaluations of common positions fairly early on in the game, then it can play better.

But having a pre-made transposition table/database of positions occurring far after the opening isn't useful, since the chance of reaching any particular position late in the game is extremely low. Unless this position is part of theory, in which case it would be in an opening book.

| improve this answer | |
  • It's a very good point about positions rarely occurring and therefore not requiring pre-calculation or storage. I'm aware the kind of database we are talking about would include billions of positions with terabytes of information. It just seems like it would be interesting to be able to instantly compare positions and see what kind of positions occur most often. I'm thinking that if we can map frequency then it will contribute to theory. – Justin Dec 14 '19 at 4:17
  • 1
    @Justin Well, the positions that occur most often are the ones which are part of theory (almost by definition). It would definitely be interesting to store other positions too, but such a task is physically impossible. The branching factor for chess is around 30, so at move X you have around 30^X positions. It grows exponentially, quickly exceeding terabytes very early on. Such a strategy would work for simpler games though. – Inertial Ignorance Dec 15 '19 at 9:02

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.