Brainfish, a new concept of a chess engine

[Ozymandias]

I've played against strong players using commercial books and Cerebellum easily holds its own and quite often out-performs the Commercial Book.
So, continue with your technical/theoretical nit-picking if it pleases you, but for all practical purposes Cerebellum is the GOODS !
I've yet to see it out-witted by ANY Commercial Book, and believe me I play a LOT of online engine-engine matches.
So, I know what I'm talking about, at least from the practical standpoint.

If you're happy with your practical results, by all means, go ahead. But at least, you know what it is and what it isn't. Other people, doing tests (not playing online) seem to be confused, about what it is that they're testing.

[jefk]

[quote="shrapnel"]
I've played against strong players using commercial books [/quote]

many of these commercial books (and those in book tournaments)
seem to be geared at -fast- blitz, with lots of games, and
book learning. For standard you need a more narrow
book and then it doesn't surprise me Cerebell is performing
well; i have a similar book with more hand tuning but i
don't release it because i'm using -partly- in correspondence
chess; which is different again; one inaccuracy, or suboptimal book
move and your lost at top-level; otherwise (most likely) draw,
whereby White usually starts with a tiny advantage.

Development of opening theory continues, and in this Cerebellum
tool there also will be no possibility (in GUI) to edit/add book moves;
normally you can add in ChessPartner a GUI book which you can
edit/adapt , and play with engine book combined (not sure
how this is done precisely, but it seems to be possible);
but as it is now, i suspect Cerebellum will override any
GUI bookmoves, which wouldn't be ideal for fast blitz or
correspondence chess. For the latter, when using minimax,
eval (at the end of opening stage) is the key; for each
engine update (stockfish, komodo) you have to generate
the whole book again completely, and i don't know how
much time this will cost (at least a few days, i suspect)
jef

[clumma]

My point was, that it's not really another engine, it's a book; sure, there's also 'large pages' in his Stockfish, similar eg as in the Stockfish clone Sugar; no big deal. Then about the book, again, it's not new, Tony Werten, with his Xinix did something like this *years* (15 yrs) ago. But his code+book was not published.

Sigh. Probably not worth responding to this nonsense, but Thomas did not claim the approach is new. Nor does anyone here think Brainfish is a new engine. It is a demonstration that a backsolved computer-generated book is a good idea. And it is already clear that all books will be generated this way in the near future. Moreover, though Werten may have worked on the same idea, there are subtitles involving repetitions etc. that make a good implementation of the idea nontrivial. If Thomas has worked these out, it will indeed be a welcome addition to the state of the art -- in contrast to some unpublished anecdote from 15 years ago.

Finally, Brainfish raises questions about stored evaluations being compiled into engines. Should it be allowed? How can it be detected (engine binary will be bigger, but maybe even small stored hashtables could give an advantage)? Is it faster than lookups through a book API?

you may call that nitpicking, but i'm also helpful to Thomas Z, warning him that if he wants to do business he better pursue commercial ambitions not mainly in computer chess but better elsewhere

Oh yes, that is very helpful.

-Carl

[fantasmadel50]

Hi everyone, I was surguieron some questions: 1) can be incorporated tablebases the gui, when it will play with brianfish and his book.?, No conflict generated? I say this because as I have understood the book also brings the end. 2) Why book light is told ,? 3) You can load the book and the brianfish engine in the sand, I wonder this arena only supports the extention abk book.? 4) The book brianfish serves to infinite analysis? 5) It is true that the engine playing with her book light is 70 elo stronger than the version without the book, or the comparison is made with the official stockfish engine.? Thank you very much in advance to anyone who can evacuate me these doubts. EXELENTE PROJECT

[jefk]

[quote="clumma"]it is already clear that all books will be generated this way in the near future.[/quote]

when i started with this concept for large books 25 yrs ago on the
website superchess.com (now taken over as url by lokasoft) and the program Bookbuilder, many people were (very) sceptical; in fact mr
Hyatt still is sceptical, tried it for Crafty but it didn't work out well;
and after all, Chessbase worked with statistics, only bookup was
the first - i admit- but did not minimax quickly large books.

Nowadays, with more chess experience, and the 2015 discussion
about the ideal opening book, i also believe some statistics
combined with the minimaxed eval can increase robustness
of the book; as for experience (Anil V Dharan) i also played fast games,
last year in the ICC monthly comp tournaments and even won a price
twice; but faster comps usually got the first price (100 $).

[quote="clumma"] (warning about commerce in computer chess)
Oh yes, that is very helpful. [/quote]

you sound as a friend of Thomas Z, but anyway, yes
my comment could be *very* helpful; i shouldn't have
written maybe the word 'warning' what i meant is friendly advice.
I've seen many come and go in trying to go commercial,
one example eg was Microvision, with their product ChessVision.
Many programming years and ofcourse he went quickly bankrupt.
jef
PS
as for the negative words by Dharan and you about my comments,
instead of a programmer, i've been inforrmation analyst at (paid)
work for a nr of years; ever heard of the socalled system developent
cycle ? Instead of just hobbyistic prototyping, professional
programmers often helped by functional analysts first do
feasibility study, then make functional specificiations,
alfa beta releas plus testing, and only then start bragging
about their product. So also by giving now my ideas to
Thomas i'm indeed helpful, and not intending to be negative.

[carldaman]

My point was, that it's not really another engine, it's a book; sure, there's also 'large pages' in his Stockfish, similar eg as in the Stockfish clone Sugar; no big deal. Then about the book, again, it's not new, Tony Werten, with his Xinix did something like this *years* (15 yrs) ago. But his code+book was not published.

Sigh. Probably not worth responding to this nonsense, but Thomas did not claim the approach is new. Nor does anyone here think Brainfish is a new engine. It is a demonstration that a backsolved computer-generated book is a good idea. And it is already clear that all books will be generated this way in the near future. Moreover, though Werten may have worked on the same idea, there are subtitles involving repetitions etc. that make a good implementation of the idea nontrivial. If Thomas has worked these out, it will indeed be a welcome addition to the state of the art -- in contrast to some unpublished anecdote from 15 years ago.

Finally, Brainfish raises questions about stored evaluations being compiled into engines. Should it be allowed? How can it be detected (engine binary will be bigger, but maybe even small stored hashtables could give an advantage)? Is it faster than lookups through a book API?

you may call that nitpicking, but i'm also helpful to Thomas Z, warning him that if he wants to do business he better pursue commercial ambitions not mainly in computer chess but better elsewhere

Oh yes, that is very helpful.

-Carl

As things stand right now, when an engine has an internal book, it will make book moves instantly. I don't think I've ever seen internal book moves that are not instant. Also, as result, there is no PV shown after such book moves.

These things (plus a large exe size, possibly) usually are enough to give away the fact that an internal book is being used.

Has anyone seen internal book moves that are disguised to look like the engine has been thinking on its own?

[Ozymandias]

Sigh. Probably not worth responding to this nonsense, but Thomas did not claim the approach is new.

This is from the very post in this thread, started by Thomas (first words after salutations):

today I released Brainfish, which realises a new concept of a chess engine,

Are you saying that "new concept" is less of a statement than "new approach"? And BTW, if no one "here think Brainfish is a new engine", it's not because of the quoted text.

If you're trying to help Thomas, you aren't doing a very good job about it.

[clumma]

you sound as a friend of Thomas Z

Never heard of him before this thread.

You sound, as several others here, like a stereotypical forum denizen, who likes to write about how everything has been "done" before, where "done" is defined as "mentioned on this forum one time". I've moderated large technical forums and I recognize it well.

Are you saying that "new concept" is less of a statement than "new approach"? And BTW, if no one "here think Brainfish is a new engine", it's not because of the quoted text.

If you're trying to help Thomas, you aren't doing a very good job about it.

I'm not trying to help Thomas, I'm pointing out gratuitous negativity and nonproductive comments, and the (obvious) fact that Thomas has built something interesting.

Engine testing wasn't a new idea, but nobody tried it at scale in the era of 64-bit machines with bitboards and PSTs before Rajlich, and that simple feat -- putting good execution behind an obvious idea at the right time -- put him on top of the engine world for five years.

Thomas did not claim Brainfish is a "new engine" in the sense Jef mentioned. He very clearly stated it is just the latest Stockfish with stored eval. So you can stop cherry picking phrases from his posts (whose native language is not English) in support of some incoherent argument you don't even really want to make.

For my part I would like to encourage Thomas to release Cerebellum. I will happily buy it if it works as claimed.

-Carl

[clumma]

Is it faster than lookups through a book API?

As things stand right now, when an engine has an internal book, it will make book moves instantly. I don't think I've ever seen internal book moves that are not instant. Also, as result, there is no PV shown after such book moves.

I meant in terms of search. Thomas has stated the database contains middlegame positions, apparently disjoint from complete opening lines. So it can function like a tablebase in terms of search. Presumably engines load opening books into memory and access is just as fast as what Brainfish is doing... but maybe not (?).

-Carl

[jefk]

[quote="clumma"]
For my part I would like to encourage Thomas to release Cerebellum
[/quote]

fine with me, in fact he can release whatever he wants;
so i'm not negative, and i think i was clear about some
-apparently technical- points in this project.
and btw it's probably not Thomas Z (*) who is
the programmer, but a younger brother, Stefan;
at least what i currently suspect.
NB also been a project leader, and yes, now
already for 25 yrs parttime in computer chess;
and yes, it has done before, and it will fail again,
if you cant combine it with other books like eg in
Chess Assistant or Aquarium; although there the
functionality is very unclear, and sometimes apparently
not even working (epd import); but i intend
to ask them that later on their forum.
jef
PS add an ICc/Fics/Chessokgaming interface to Aquarium
and you got almost all functionality (except that cerebellum
might be faster); also i would recommend Chessok
to make common directories in 'my documents'
for games/trees etc just like chessbase is doing. Currently
it's a mess, if you try to combine Ca and Aquar.
PS2 (*) so if i'm right that thomas Z is the commercial
guy then indeed like juan molina pointed out he
knows zip about computer chess. but clumma if
you want to buy it, fine with me. others- if it will work-
probably will grab a copy, and you, just like Thomas Z
apparently have not clue about the amount of piracy
in chess; most above IM level know but i'm not going
to betray them (just a hint, they copy everything;
chances of becoming a commercial succes are slim,
unless you are happy to program for eg 40 cts/hr
which people in Banglah Desh would not even accept