My Deep Rybka 4 SSE42 x64 just drew a game against a much weaker engine. This can happen once in a while and it is not what I am complaining about. This version was up a (doubled!) pawn, in a Rook and Pawn endgame, however, the rook was in front of the pawn, making the position a dead draw. In spite of this, Rybka gave a winning score for a long time. In the diagrammed position, Rybka shows +2.43.
C'mon Vas, how hard is it to fix this eval bug that you are aware of?!
It's not like, a piece of cake fixing bugs. I don't think Vas wants to add exceptions to every single complaint he gets on the forum. Rather, you want to discover a pattern in the position to address the entire issue, instead of just fixing up a specific bug. So I don't know, maybe Vas will add doubled-pawn-in-endgame-and-bad-rook-placing when he finds a good way to do it.
Oops Stockfish is based on Glaurung, so, it have some of the code from it, right? so, some of the code is cloned.
The term clone is derived from κλῶνος, the Greek word for "trunk, branch" <- I just cloned the line from wikipedia.
Stockfish scores the position as +2.x.
If you want to know the back story, the open-source Glaurung engine developed by Tord was kind of "put on hold" when Tord went to pursue other interests. Marco came along and made various improvements to Glaurung and named it Stockfish, also released under GPL. In that sense it isn't really a "clone," which is a word with negative connotations, but rather a "fork," or divulgence from the path.
>C'mon Vas, how hard is it to fix this eval bug that you are aware of?!
Generally, very hard. Quite often, fixing these kind of things makes the ELO go down, because there are about 10 non-obvious extra conditions that have to be true before it should be activated, and you don't think of them.
It's easy to point to a single position where an engine goes wrong, but to "fix" a position and actually improve the engine you have to cover all other possible positions too.
> You can thank the Beta testers for this one, they wouldn't let me sleep until I added it. :-)
I hope the next beta testers won't let you sleep until you include bishop under-promotion. Without it, Rybka does not fully understand chess :)
> I agree. BU irrelevant.
I once played a game which needed BU to avoid stalemate (Queen or Rook we stalemating) and it was delivering a quick mate. I was very excited to see that. My opponent resigned immediately when he saw me under promote to a bishop :)
> You actually played a game where BU was important??? :-)
Bingo! You'll never know when you gonna need it :)
I also got one remarkable game: my computer announced a mate in 91 - it ended in a won knnkp position :)
Rybka happily underpromots to a knight it seems when necessary, it should be able to do the same with a bishop.
>> Rybka happily underpromots to a knight it seems when necessary, it should be able to do the same with a bishop.
There's a huge difference! The knight can do things the queen cannot which makes it uniquely suitable for promotion in certain types of positions. Positions where a queen promotion leads to a worse position than a bishop promotion are exceedingly rare.
I remember him because in this offhand game, I was White, and, incredibly, and without any irony, the game went 1 e4 g5 2 d4 f6 and you may be able to guess the next move without using Rybka.
Never thought it would ever happen, but it did.
A propos of I'm not sure what...
or a UCI switch, and be done with it!
For analysis what do you think the best setting for Rook Endgame Scaling is?
I have the complete 3 and 4 man tablebases on my computer.
The more I have seen R4 games the more I have toned down my criticism of its endgame play, and I think I now understand why Vas has left it as it is. R4 wins some pretty amazing games by simply outplaying its opponent in rook endgames. The overly high evaluation makes R4 be quite aggressive and dynamic, and in many cases it will squeeze out a win. It is most likely the best compromise for R4 at the moment. Of course it is hugely frustrating when R4 avoids a winning endgame to simplify into an obvious rook endgame, but if you look at an endgame book you will see that 3/4 of it is only about rook endgames. That is not unusual since 10% to 15% of games end up in rook endgames, so it is significant. I am not sure how much 3/4 of a book of heuristics is in chess code, but most likely it is a major project.
By the way, rook endgame play by R4 can be greatly enhanced by having the relevant 6 man EGTB's. I am sure you can find info on that if you do a search. Am not sure how many GB the relevant rook EGTB's are, but they should be reasonably manageable.
I definitely won't mess with Rook Scaling for game play. (that is a small part of how I use Rybka anyway)
But I will definitely use your setting of 60 for Rook Scaling when doing analysis (which is the biggest use I have for Rybka)
I have also been thinking of using Rybka 4 as my engine for practicing endgames (pgns etc.). I have been using Zappa Mexico II but every once in awhile I have noticed that Rybka might play harder for the draw in a losing situation so that would be a better test for me!! Thanks again for the help!
> I hope the next beta testers won't let you sleep until you include bishop under-promotion. Without it, Rybka does not fully understand chess
With it too. :)
> With it too. :-)
Yeah sure :-D
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill