TalkChess.com

Calling API over the Internet for learning at any meaningful scale is not efficient, you should then set up local environment from my data snapshots and make changes to API as you see fit.

Would it be possible to let LC0 use the full 7 piece tablebases to do 7 piece rescaling in training?

A new snapshot has been made available for download which includes all evaluation data(about 10 billion unique positions) as of yesterday.

noobpwnftw wrote: ↑Thu Jul 09, 2020 10:37 pm A new snapshot has been made available for download which includes all evaluation data(about 10 billion unique positions) as of yesterday.

Hi noobpwnftw, could you share a sample of the snapshot of ftp://ftp.chessdb.cn/pub/chessdb/data-s ... 200708.tar? The idea is to check contents and perform some quick tests. 100MB in size would be great!
In addition, is there a description of the contents of the db? I mean, just opening theory, or some middlegame/endings too.

Thanks in advance, Carlos

carangelmx wrote: ↑Thu Sep 03, 2020 5:56 pm In addition, is there a description of the contents of the db? I mean, just opening theory, or some middlegame/endings too.

Any position with 5 or more non-pawn non-king on the board can be evaluated by ChessDB.

Nice tool for chess enthusiasts and masters alike, thanks for the info.
I tried the difficult Q sacrifice on DBCN. I was expecting to see the solution, but doesn't show the correct move.

bm is Qxe5


First of all, complete game is available online.
Second, this has been analyzed to death in almost all forums.
How come the winning move Qxe5! was not listed at the top of possible continuations?

MikeGL wrote: ↑Sun Sep 27, 2020 1:43 am Nice tool for chess enthusiasts and masters alike, thanks for the info.
I tried the difficult Q sacrifice on DBCN. I was expecting to see the solution, but doesn't show the correct move.

bm is Qxe5


First of all, complete game is available online.
Second, this has been analyzed to death in almost all forums.
How come the winning move Qxe5! was not listed at the top of possible continuations?

No one has entered it yet. The database is made by Stockfish continuosly exploring lines from the opening, it's not looking at mid or endgame puzzles on its own. You can add stuff manually by requesting analysis, and you can add lines showing the way to win in difficult puzzles by picking the right moves, requesting analysis along the way, like I've done with this one:
3rrk2/1bnqppb1/p2p2p1/1p1P2P1/1p2PP1R/2NQBN2/1PP2K2/7R w - -

On first post in 1st page of this thread, I thought DBCN is crawling the web and forums and saving all new positions it encounters including the move, then storing it into DBCN.

But ok your post clarifies the issue on why it was not in DBCN,
Thanks for your reply.

MikeGL wrote: ↑Sun Sep 27, 2020 1:56 am On first post in 1st page of this thread, I thought DBCN is crawling the web and forums and saving all new positions it encounters including the move, then storing it into DBCN.

But ok your post clarifies the issue on why it was not in DBCN,
Thanks for your reply.

I explored some lines from your position, it now gives Qxe4 as the top move, but I didn't explore enough to show it is winning, so the score given is 0

Terje wrote: ↑Sun Sep 27, 2020 2:36 am

MikeGL wrote: ↑Sun Sep 27, 2020 1:56 am On first post in 1st page of this thread, I thought DBCN is crawling the web and forums and saving all new positions it encounters including the move, then storing it into DBCN.

But ok your post clarifies the issue on why it was not in DBCN,
Thanks for your reply.

I explored some lines from your position, it now gives Qxe4 as the top move, but I didn't explore enough to show it is winning, so the score given is 0

Ok thanks.
Revisited the first post again, I apologize for missing the point in some of it. The probing mentioned there was the client (user) of ChessDBCN who is trying to probe so many positions at ChessDB, and not ChessDB site crawling the web and forums for new chess positions. That would make it a hyper AI, i guess