Brainfish, a new concept of a chess engine

[Dann Corbit]

It's an indication of popularity.

Exactly my point, it only proves that marketing is working. It doesn't say anything about the underlying quality of the product being marketed.

BTW, let me be clear, the product being promoted is NOT what's being developed. The software intended for selling, is just a part of the package being given as a sample.

Still, it is an interesting idea. Not that it is novel, the fundamental idea has been brought up many times.

I believe that MVK did a minimax of his data pile several years ago.

[Ozymandias]

it is an interesting idea. Not that it is novel, the fundamental idea has been brought up many times.

I believe that MVK did a minimax of his data pile several years ago.

A similar tool has been available already for quite some time, commercially. Both Chess Openings Wizard Professional as well as ChessOK Aquarium, offer back-solving and iDea respectively.
The only novelty, from a developing standpoint, is the presence of an adapter that allows an engine to take profit of the tree generated. This makes it possible, for a marketing proof of concept to exist.
Not surprisingly, just that was all you needed, for lots of people to flock to BrainFish. I think the aforementioned products missed an opportunity.

[reflectionofpower]

It would be a nice concept if you could download it. Every link I tried to download it from said it could not be reached?

[Eelco de Groot]

It would be a nice concept if you could download it. Every link I tried to download it from said it could not be reached?

Thomas wrote here he is making a new version, where the book is an external file. Should be ready around Friday.

[shrapnel]

Good that I had already downloaded it
Go, Thomas, Go , you have a Winner !
By the way, I'd be happy to be a Tester for you, though I only have a humble 8-Core i7 5960 X with 64 GB RAM ; though I've beaten a few Dual-Xeons and drawn quite a few, on Infinity Chess.

[Dann Corbit]

it is an interesting idea. Not that it is novel, the fundamental idea has been brought up many times.

I believe that MVK did a minimax of his data pile several years ago.

A similar tool has been available already for quite some time, commercially. Both Chess Openings Wizard Professional as well as ChessOK Aquarium, offer back-solving and iDea respectively.
The only novelty, from a developing standpoint, is the presence of an adapter that allows an engine to take profit of the tree generated. This makes it possible, for a marketing proof of concept to exist.
Not surprisingly, just that was all you needed, for lots of people to flock to BrainFish. I think the aforementioned products missed an opportunity.

Bookup also has a minimax type function. But again, did not give direct access to the chess engine.

[jefk]

well indeed minimax is nothing new; but sure Brainfish
seems to give quite a decent/good book (although it's not
clear for me how it's book learning works), but it's not the
only recent effort in this area, Ed Schroder last year
also worked on such topics: see eg :
http://rebel13.nl/prodeo/index.html
http://rebel13.nl/misc/ebf/protools.html

Sofar most opening books still are in GUI format; in future i
think we need better standardization (considering that polyglot
has not so much features as a 'perfect' book would need and
thus also is not the ideal solution). For your (and the Brainfish
author's) info, last year there was a long discussion (thread) on
opening books, what came out as theoretical recommendation
was a mix of opening position backsolving/minimax and
statisticis; see eg here :
http://talkchess.com/forum/viewtopic.ph ... 31&t=55569
and here:
http://talkchess.com/forum/viewtopic.ph ... 17&t=55569

So ideally in a certain position, the 'best' book
move to play would follow from an algorithm comparing
the bookmoves according to eg 'value' + 'c'(posstat-50)/100
with 'value' the minimaxed openingposition value (eg 0.12 pawn)
and posstat the statistical results in a Gm/compgames base
eg 63.3 = 63.3 pct (with the usual b/w/ chessbase calculation)
with c, preferably also taking into account statistics reliability
eg (1 - 1/nrofgames); a new element, which i now propose..

A bit technical the above i admit, so maybe i should rewrite and
copy it in the subforum here about compchess *programming*.
Because of the statistics element the program would also
need a database, and then it becomes like a GUi again..

Chessbase with their livebook could already work on
such a system, (averaging the positional score in the
lets check analysis for various engines, and then also
use some statistics eg from the live book or other) but
apparenty sofar they do not seem to be interested in such
'theoretical' computerchesstechnicalities unless it would
have commercial benefits i suppose.. So next candidate
then might be Arena, but progress there seems slow
in recent years, is my impression.

ah well, in the meantime i consider running some testgames
probably rapid time control with a big Schroder book on
Arena vs Brainfish, and my personal ChessPartner book
vs Brainfish, and maybe inform you about the results
(it will be anyway some good games to load into my CP
book again, keeping things /opeing moves up to date..).

jef

PS nowadays with top engines i think you can create
quite decent books compared with GM books; the best
opening moves however are played in top correspondence
chess, eg on ICCF, where also computers can be used.

PS2 [quote="Ozymandias"]
A similar tool has been available already for quite some time, commercially. Both Chess Openings Wizard Professional as well as ChessOK Aquarium, offer back-solving and iDea respectively[/quote]

yes, and Bookbuilder ofcourse www.bookbuilder.nl
Recently updated the book, now on basis of Komodo10,
and comparing with Brainfish in some variations (eg
Sicilian Najdorf) it looks to me i got better results.

[Ozymandias]

Bookup also has a minimax type function. But again, did not give direct access to the chess engine.

It shouldn't, bookup was the predecessor to Chess Openings Wizard

yes, and Bookbinder ofcourse www.bookbuilder.nl

I didn't know about this one:

With an 'interface' (eg. Winboard) you can play with a 'normal' chess
program (e.g. Crafty), using the Bookbinder generated book as an advantage

So, it looks like this one, does everything the Cerebellum is offering?

[jefk]

[quote="Ozymandias"] (Bookbuilder) With an 'interface' (eg. Winboard) you can play with a 'normal' chess program (e.g. Crafty), using the Bookbinder generated book as an advantage[/quote] So, it looks like this one, does everything the Cerebellum is offering?[/quote]

uh, well not everything anymore; in early years i used it to
generate a crafty book(.bin) and indeed played many testgames;
but nowadays with the UCI engines, i started to use ChessPartner
(from Lokasoft) and making such a book was a different exercise.

However i still often compared the lines from the CP book (.bk format)
with my large Epd posbase (.pbs) made with bookbuilder,
and with such crosschecks, also sometimes looking at GM (paper)
books about opening theory, gradually improved the book.

A few years ago, Chessbase sold a few (.ctg) books Dufek,
Noomen, and so on. Nowadays there are commercial books
from Hiarcs (.ctg), Fauzi (.abk) and 1337,
http://1337chess.blogspot.nl/p/blog-page.html
they all claim to have the best book ofcourse ..

Anyway, besides the Bookbuilder program which is
a bit for specialists, the EPD standard is well known
and in principle it should be possible to load an EPD
book into Aquarium, optimize it, (ideally they should
use the formula i described) and then use it in ChessAssistant
for online play. This however is cumbersome..

jef

PS sofar i haven't even been able to load my Epd book
into Aquarium, and perform a full mima, but i have to look
deeper into the documentation and do a few more tests
to see if it works.
PS2 did a brief test Brainfish in Arena vs Prodeo Yat,
but apparently the second move probability in Brainfish
didn't work and the games were always the same,
Italian and RL Marshall, both with drawish outcomes..
Maybe Friday we'll see a bit more from the Brainfisht dpt..

[Thomas Zipproth]

Would your tool also work for making min-Shogi books?

From what I read about mini-.Shogi, the Algorithm would work, if you have a score for every position you want to include.
Of course there is no move generator which can be simply included, so it would mean some work to do that.

Don't listen to the nit-pickers and cribbers, Thomas.
Your Brainfish is excellent.
Nearly 50 % of online gamers are using your Engine, much more than Komodo !

Thanks for your encouragement, the new Version with external book has arrived. So more regular updates of the book are possible.