On-line engine blitz tourney December

Discussion of chess software programming and technical issues.

Moderators: hgm, Dann Corbit, Harvey Williamson

Joost Buijs
Posts: 1562
Joined: Thu Jul 16, 2009 10:47 am
Location: Almere, The Netherlands

Re: On-line engine blitz tourney December

Post by Joost Buijs »

Pi4Chess wrote: Sat Dec 12, 2020 11:51 pm
jshriver wrote: Sat Dec 12, 2020 11:34 pm
Pi4Chess wrote: Sat Dec 12, 2020 10:27 pm -The book that won the last Sedat chess books tournament
Very impressive. Do you know how to convert the book to pgn or polyglot? Seems the author released it only in chessbase format.

https://sites.google.com/site/computers ... rfect2020b
It's not this book. This book is from Sedat himself but the book that won the last tournament is Perfect by Cevdet. It might be cerebellum3merge with lines updated with last top engines games/evaluations.
i picked it at Chess2u forum in polyglot format ;-) I don't think it is good for big configs but for little ones it helps :-)
Cevdet-cerebelium is indeed a very strong book. Before the tournament started I let my engine play some games against your PI4, and it happened several times that my engine got out of book in a (more or less) lost position. With 127MB it's a lot smaller than the Cerebellum3merge, maybe it is based on an older Cerebellum book.

I will add NNUE too, the difference between handcrafted and NN evaluation is getting to big. I've already done some tests with NNUE but it is not ready for prime time yet. I think the next tourney will be over 4 weeks (depends upon the TD), hopefully by then I will have a NNUE implementation of which I'm satisfied.
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: On-line engine blitz tourney December

Post by Pi4Chess »

Joost Buijs wrote: Sun Dec 13, 2020 7:58 am
Cevdet-cerebelium is indeed a very strong book. Before the tournament started I let my engine play some games against your PI4, and it happened several times that my engine got out of book in a (more or less) lost position. With 127MB it's a lot smaller than the Cerebellum3merge, maybe it is based on an older Cerebellum book.

I will add NNUE too, the difference between handcrafted and NN evaluation is getting to big. I've already done some tests with NNUE but it is not ready for prime time yet. I think the next tourney will be over 4 weeks (depends upon the TD), hopefully by then I will have a NNUE implementation of which I'm satisfied.
All the games are saved so I will look at them. What you telling is the exact situation WaspX experimented during tournament :at move 15 leaving opening book it was already +1 score for Cfish and +3 after few moves.
On the other hand rpiRofChade kept the game in an inextricable drawish position when he was playing black.

As for nnue implementation, if it's SF one of course you might win some elo. The trick is like SF using it in hybrid mode and decide when it has to use classical or nnue evaluation. Well i am just an enthusiast user while you are engine authors so it's quite different and "my" performance with pi4 is just an assembler one.
Pi4Chess
Posts: 253
Joined: Mon Nov 16, 2020 12:13 pm
Full name: Manuel Rivera

Re: On-line engine blitz tourney December

Post by Pi4Chess »

Well my memories were not accurate.

It's marvin with black that is in bad position when rpiforchess leaves book very early : [Annotator "12. +1,02"].
WaspX with black is lost when rpichess leaves book at move [Annotator "27. +4,07"]
rpiRofChade was playing white when draw game.