TalkChess.com

Incredible Stockfish patch by Joost VandeVondele

LLR: 2.96 (-2.94,2.94) [0.00,5.00]
Total: 4093 W: 571 L: 441 D: 3081 sprt @ 60+0.6 th 1 ltc, lp-- variant

Congratulations to all Stockfish team


http://tests.stockfishchess.org/tests

Great news, but...

what a heck means all that jargon?

Fern

Instead of always searching on the traditional alpha-beta window of -Infinity to Infinity (IE, looking for any move between -Inf and +Inf), most engines will use something called aspiration windows.

If I believe I am currently up by one pawn, I will search on a window of (.75Pawns , 1.25Pawns). This allows AlphaBeta to produce more cutoffs. I choose this window because I don't expect any drastic change in the value of my position.

Traditionally, when AlphaBeta returns a result outside of the window, you will perform a new search with a larger window. I do this, Stock fish currently (although maybe not for long) does this, and as does every engine I have ever looked through the source of.

What Joost has done is in the event you get a result above the window (Meaning the move was better than you expected ANY move to possibly be), and this move is what we thought was best on the last search, we will opt to skip the research.

This could have some poor affects for analysis, Instead of Stockfish giving a few scores for each depth, as it zeros in on a true value, it will exit a depth early with a lowerbound score of a move.

Alternatively, instead of wasting time searching a move over and over again which you expect to be best, just to get a more refined value, Stockfish will move onto search the move deeper. I'll be interested to see what those who use Stockfish for analysis think.

So with that technique you win time?

Sorry if I look as an idiot on this, but I am not computer expert and perhaps I am an idiot too.

Fern

AndrewGrant wrote:This could have some poor affects for analysis, Instead of Stockfish giving a few scores for each depth, as it zeros in on a true value, it will exit a depth early with a lowerbound score of a move.

Actually, I've seen Komodo do this (move to the next depth on a fail high) and it doesn't have a poor effect on analysis, because when the evaluations are jumping like this, the user is best making the move and reaching more relative depth, to see if the fail high persists, anyway.

There's no such thing as a "true value", so I think this search strategy makes sense.

Someone should put a learn file inside Stockfish,at least make one with that option and call it StockfishL series for Stockfish Learn.Make it able to access lots of hash as needed.


FWCC

FWCC wrote:Someone should put a learn file inside Stockfish,at least make one with that option and call it StockfishL series for Stockfish Learn.Make it able to access lots of hash as needed.


FWCC

It was done 3 years ago:

http://www.open-chess.org/viewtopic.php?f=7&t=2663

Jeremy Bernstein wanted to keep the latest Stockfish with learning, until the Komodo team invited him to join them, so he switched sides.

Ovyron wrote:

FWCC wrote:Someone should put a learn file inside Stockfish,at least make one with that option and call it StockfishL series for Stockfish Learn.Make it able to access lots of hash as needed.


FWCC

It was done 3 years ago:

http://www.open-chess.org/viewtopic.php?f=7&t=2663

Jeremy Bernstein wanted to keep the latest Stockfish with learning, until the Komodo team invited him to join them, so he switched sides.

I do not think Jeremy has "sides". He modified cutechess-cli for TCEC, added Syzygy to Komodo did work on Stockfish, like you mention. He also created open-chess.org computer chess forum. Basically, Jeremy is awesome and promoted computer chess to everyone.

Mark

mjlef wrote:

Ovyron wrote:

FWCC wrote:Someone should put a learn file inside Stockfish,at least make one with that option and call it StockfishL series for Stockfish Learn.Make it able to access lots of hash as needed.


FWCC

It was done 3 years ago:

http://www.open-chess.org/viewtopic.php?f=7&t=2663

Jeremy Bernstein wanted to keep the latest Stockfish with learning, until the Komodo team invited him to join them, so he switched sides.

I do not think Jeremy has "sides". He modified cutechess-cli for TCEC, added Syzygy to Komodo did work on Stockfish, like you mention. He also created open-chess.org computer chess forum. Basically, Jeremy is awesome and promoted computer chess to everyone.

Mark

Komodo with such a learning feature would be a much better analysis tool. Stockfish and Houdini would be worth much less in comparison, if that were to happen.

Cheers,
CL

Thank you,but it's not in the latest version.Why was it stopped and can a user add it.What were the results?