GNU Chess 5.07.175b released

[Michel]

Hi, I put GNU Chess 5.07.175b at the usual place

http://hardy.uhasselt.be/Toga/gnuchess-release/

The .zip file contains the source as well as binaries for linux32/64, windows32/64 and android. Of course these have only been tested on my own computer (the foreign binaries with wine and qemu-arm).

This version of GNU Chess appears to be of similar strength as Toga 1.4.1SE at 1m+1s time controls (confirmed by 5000 games).

Here are the changes that gave Elo gain at fast time controls (as far as I remember)

Use SEESign instead of the less efficient SEE.
Do futility pruning without a MakeMove (this is a no brainer of course).
SEE pruning (a very nice idea of Lucas, now also used in Stockfish).
History ageing.
Rudimentary implementation of KRPKR.
Free passer bonus.
CLOP tuning of passed pawns.
CLOP tuning of aspiration search margins.
CLOP tuning of king PST.
CLOP tuning of delta margin.

[michiguel]

Hi, I put GNU Chess 5.07.175b at the usual place

http://hardy.uhasselt.be/Toga/gnuchess-release/

The .zip file contains the source as well as binaries for linux32/64, windows32/64 and android. Of course these have only been tested on my own computer (the foreign binaries with wine and qemu-arm).

This version of GNU Chess appears to be of similar strength as Toga 1.4.1SE at 1m+1s time controls (confirmed by 5000 games).

Here are the changes that gave Elo gain at fast time controls (as far as I remember)

Use SEESign instead of the less efficient SEE.
Do futility pruning without a MakeMove (this is a no brainer of course).
SEE pruning (a very nice idea of Lucas, now also used in Stockfish).
History ageing.
Rudimentary implementation of KRPKR.
Free passer bonus.
CLOP tuning of passed pawns.
CLOP tuning of aspiration search margins.
CLOP tuning of king PST.
CLOP tuning of delta margin.

What is SEESign?
What do you call SEE pruning?

You have done a lot work and revived this engine. At this point you should make GNUChess your own and fork it. The natural option should have been to make it GNUChess 6.0, but since it was decided by the GNU people to "clone" Fruit 2.1 instead for the 6.0 version, you can't do it. At least you call it GNUChess 5.5 (or 7.0 )

Miguel

[Michel]

What is SEESign?
What do you call SEE pruning?

It is faster to determine if the SEE value is below some threshold than computing its actual value.

"SEE pruning" is pruning moves with negative SEE (for futility purposes). This is commonly done in quiescence search but it turns out you can do it in the main search as well.

You have done a lot work and revived this engine. At this point you should make GNUChess your own and fork it. The natural option should have been to make it GNUChess 6.0, but since it was decided by the GNU people to "clone" Fruit 2.1 instead for the 6.0 version, you can't do it. At least you call it GNUChess 5.5 (or 7.0 )

Miguel

Thanks for you kind words. GNU Chess 5.5 sounds nice although I have problems with the advertizing aspect of engine names and version numbers. Perhaps something for the next version, which might take a while.

Michel

[mar]

Thanks and congratulations, Michel!
Just a remark - wouldn't this announcement be more visible in the general forum?

[lucasart]

SEE pruning (a very nice idea of Lucas, now also used in Stockfish).

Don't credit me for SEE pruning. It was in Glaurung as far as I can remember. So you can credit Tord Romstad. But I'm sure that others did it before him. It's a logical thing to do.

What is perhaps less common in DiscoCheck's qsearch are:
- depth limit in the QS, after which (if not in check) I return eval+see
- never do SEE pruning (in search or qsearch) in the case of a disco check, where the see is considered meaningless.

Other than that, what I'm doing is basically standard.

[Michel]

Don't credit me for SEE pruning. It was in Glaurung as far as I can remember. So you can credit Tord Romstad. But I'm sure that others did it before him. It's a logical thing to do.

Oops, sorry about that! I saw your name on some of the SEE pruning patches in the Stockfish test queue and assumed the whole thing was your idea.
With hindsight I guess it is indeed a logical thing to try.

[Michel]

Just a remark - wouldn't this announcement be more visible in the general forum?

Yes I guess the general forum would have been more appropriate. As I read both forums I don't generally make the distinction.

[mcostalba]

SEE pruning (a very nice idea of Lucas, now also used in Stockfish).

Don't credit me for SEE pruning. It was in Glaurung as far as I can remember. So you can credit Tord Romstad. But I'm sure that others did it before him. It's a logical thing to do.

The original see pruning in qsearch is from Tord since Glaurung times. The use of see pruning in main search is from Joona and was not in Glaurung.

[Michel]

The original see pruning in qsearch is from Tord since Glaurung times. The use of see pruning in main search is from Joona and was not in Glaurung.

Thanks again for clarifying. I'll be more careful next time.

[jd1]

Hi, I put GNU Chess 5.07.175b at the usual place

http://hardy.uhasselt.be/Toga/gnuchess-release/

The .zip file contains the source as well as binaries for linux32/64, windows32/64 and android. Of course these have only been tested on my own computer (the foreign binaries with wine and qemu-arm).

This version of GNU Chess appears to be of similar strength as Toga 1.4.1SE at 1m+1s time controls (confirmed by 5000 games).

Here are the changes that gave Elo gain at fast time controls (as far as I remember)

Use SEESign instead of the less efficient SEE.
Do futility pruning without a MakeMove (this is a no brainer of course).
SEE pruning (a very nice idea of Lucas, now also used in Stockfish).
History ageing.
Rudimentary implementation of KRPKR.
Free passer bonus.
CLOP tuning of passed pawns.
CLOP tuning of aspiration search margins.
CLOP tuning of king PST.
CLOP tuning of delta margin.

Very very impressive! I agree, you should rename it (I suggest GNUChess 7.0 as it is much stronger than GNUChess 6) and publicize the release in the general forum so the testers know about it.

Jerry