Willow 4.0 (final release for now)

[Frank Quisinsky]

After all ... I often think that the pawn structure is the secret of very strong chess engines.
I get such impressions again and again when I look at Seer games. Seer is not an attacker, but the pawn structures are always good. The higher the time control, the higher the boost Seer gets with such a good pawn structure.

And after all the statistics I create here ... if the day is long, you know ...
Seer's pawn structure is better than Uralochka's.

When such a strong engine as Uralochka (a wonder in attacking chess) has in combination the pawn structure of Seer ... Stockfisch is really in danger of losing 1st place in computer chess.

Back to your engine:
Very strong what your engine is playing here in Blitz
https://www.amateurschach.de/fling/test-39.html

Only the results against Wasp are bad, i.e. from Wasp's point of view.



Best regards
Frank

John is working!
Enough ... have a nice day and good luck with your ideas.
I will follow your developments and thank you for your time!

[pohl4711]

I know about the EAS tool but I am trying to find out what aggressive positions look like, not just aggressive games.

As far as I understand, this is exactly what Ed Schroeders tool does
https://rebel7775.wixsite.com/rebel/kop ... rl-blitz-1

Also, the fact that Stockfish and Torch are at the top by a country mile suggests that a large part of what EAS is measuring is engines taking advantage of tactical mistakes by other engines rather than actively seeking out an aggressive play style.

That is false. If you look on my old SPCC-ratinglist and the EAS-ratinglist, calculated out of it:
https://www.sp-cc.de/files/spcc_full_list.txt

You will see, that Stockfish 15.1 and Stockfish 16 are only on rank 51 and 52 in the EAS-ratinglist (!!!)
And both have (of course) scored more than 67% vs. their opponents. Because in the testruns, all opponents were weaker than SF 15.1/SF16
And the first place in the EAS-ratinglist (Velvet 4.1.0) is not only much weaker (of course), but also have an overall score of only 53.1% vs. its opponents. Because Velvet 4.1.0 was tested vs weaker and stronger opponents...

So, the EAS-score does not depend on high scoring-percents. But on aggressive play.

The only reason, why Stockfish and Torch are so good in the UHO-Top15 EAS-Ratinglist, is, that they are playing more aggressive than the other engines in this Top15-Ratinglist. Sadly, right now, really aggressive playing engines like Velvet or so are just too weak, to enter the UHO-Top15 ratinglist.
And in the UHO-Top15 EAS-ratinglist, Rebel EAS is just behind Stockfish in the categories most sacs overall and most 3+ high-sacs. And not Torch.
And Rebel EAS has a scoring of only 37.1% and is only on rank 14 in the UHO-Top15 ratinglist... (the EAS-tool delivers much more useful stats, than only an EAS-score and a list, built out of this score.)

Code: Select all

A: Most high-value sacrifices (3+ pawnunits): [1]:04.29% Stockfish 240126 avx2   
                                              [2]:03.82% Stockfish 16 230630   
                                              [3]:02.68% Rebel EAS avx2   
                                              [4]:02.33% Torch 1 popavx2   
                                              [5]:02.11% Uralochka 3.40a avx2  


B: Most sacrifices overall                  : [1]:22.12% Stockfish 240126 avx2   
                                              [2]:21.07% Stockfish 16 230630   
                                              [3]:19.26% Rebel EAS avx2   
                                              [4]:16.15% Torch 1 popavx2   
                                              [5]:15.53% KomodoDragon 3.3 avx2  

Or (finally) look on Ed Schroeders website:
REBEL-EAS is an in between version based on the same, but more, neural net data as REBEL-16.2 but heavily modified during the last 5 months in order to make REBEL to play more aggressive and runs as a module under Chess System Tal 2.0
REBEL 16.2 gains an EAS score of : 80.258
REBEL-EAS gets an EAS score of : 160.569

https://rebel7775.wixsite.com/rebel
(And Ed tested the different Rebels vs. same opponents and Rebel EAS is only slightly stronger than Rebel 16.2 (around +25 Elo or so), but the EAS score doubled(!))

[peter]

I have Windows 10 as well and I also get messages about it being a potentially dangerous download when I try to download. This is a problem that has also randomly happened with some of my engine dev friends. If it's not letting you download the executable directly, you can build Willow from source using the instructions provided in the release notes.

Also, v3 is faster than v2 and should work on most machines. Only use v2 if you know v3 won't work or have already tried and failed to use v3.

The message of pontentially dangerous download I omitted but then the Trojan of the exact name I wrote appeared and the file had to be deleted or quarantined.
In meantime I took the bmi2- compile from Dariusz' Chessengeria- site with warning at download too but no more objection of Defender explicitly checking the already stored files again.
Engine seems not to accept exact time per move- requirement, e.g. in Shredder (and doesn't work neither with Ferdinand Mosca's MEA- tool and it's kind of time per move- setting by batch-file and command- line) yet works as for other demands I tried so far in Shredder (13)- GUI. In Fritz (19) it freezes with start- stop- start of analysis- mode, regards

[Dariusz]

Adam, thank you beautifully

I compiled Willow 4.0 for Linux, Mac and Windows - downloadable from my site (Files section).

Could you update them with the newest release please? The info string still says "Willow 3.1" because of an oversight that has since been fixed.

Sure, I did it before compilations