BrainFish vs book Chessbasse

[velmarin]

A little experiment.
BrainFish vs Stockfish with book. ctg.
The first has been to find some lines from the brainy book, so have run several races of 10 seconds between Brainfish against brainfish,
Not much, some 500 games, with these games have prepared the book muscle.ctg,
Now various matchs between last Stockfish against Brainfish, stopping occasionally to feed with the games the book.
The book is without "bookmaker" bone without manually touching only with the evaluations made by the Fritz GUI.
Only 1260 games, some with multiple passes, have been imported the book has only 7 megabytes in size.

Finally the match, 62 games of 1 minute.
Result

DESKTOP-QFSOBQ2, Blitz 1m 2016
1 Stockfish 020816 64 +22 +9/=48/-5 53.23% 33.0/62
2 Brainfish 1 64 ....... -22 +5/=48/-9 46.77% 29.0/62

PGNs, games of the match and the book.
http://www.mediafire.com/download/572y8 ... n/pgns.zip

Book muscle.ctg
http://www.mediafire.com/download/u6b7y ... muscle.zip

[Ferdy]

A little experiment.
BrainFish vs Stockfish with book. ctg.
The first has been to find some lines from the brainy book, so have run several races of 10 seconds between Brainfish against brainfish,
Not much, some 500 games, with these games have prepared the book muscle.ctg,

I got this so far.

Now various matchs between last Stockfish against Brainfish, stopping occasionally to feed with the games the book.

You stop the match and add the finished games to muscle.ctg?

The book is without "bookmaker" bone without manually touching only with the evaluations made by the Fritz GUI.

I can't understand this, can you explain it in a different way?

[velmarin]

To find the lines of Brainfish first, a BrainF match vs BrainF, you can toggle the "Book Move2 Probabilyte" this would be, 0 and 0, 0 and 100, 100 and 100.
Suite of openings can be used, but they must have all the plays, not only the FEN. "e4, c5, Nf3, d6, d4 ect," from the beginning, the book BrainF not acts of another way.
How many more games and more suites, more lines found.
In a moment made copied do the book. Copied the games,
I do "relative length" 140 a bit exaggerated, but then lose data within the book...
Is can make a pair of imports of those same games,
or three, or four, this is for the statistics.



Now those games between BrainF and muscle_book, this takes more lines and complete the statistics.
Many more, better, they must play them without "opening suite" f
rom otherwise BrainF work,
but the GUI will reject the CTG book. No work.
We import from time to time, so the book is will be completing.

Now we have the book without "Touchup" how I've left muscle.ctg:


Fritz GUI, the moves mark with right-click of the mouse, if we put them how "played in the tournament, will be green and will necessarily be played.
If them mark how "not is played of tournament" is will return red and not is will play in any way.
so would be:

The plays that are so "percentages in red and green" are modified by Fritz statistically according to the results of the game. ""
"a smart way, or brain..."?


So far, can do much more, or otherwise.
This was an entertainment for a couple of hours.

[Ferdy]

Did you manually make some moves green and red?

[velmarin]

Not the book played as well.


Only has been a small experiment by my part.
I think that Cerebellum may be a good tool, time will tell it,
How you manipulate the games, how is optimized, ect.
but the strength of Chessbasse GUIs, it cannot be denied.

[Thomas Zipproth]

This was indeed an interesting experiment, I will try to comment on it.
Let me state first because there where some emotions in other threads about that topic, that I don't want to defend or promote something, I only like to discusss it in a scientific manner.

What was done in principal in that experiment?

1.) You played a lot of Brainfish-Brainfish games, therefore basically copying the lines into a .ctg Book.
2.) You noticed all lines that where lost for black or white under the very short time control of 1 minute.
3.) You replayed the winning lines against BrainFish and of course you won many of that games again.

But what does that mean, is it really so easy?
I think it is not, I will try to explain why.

The first point, but not the important one is:

This book is not made for 1 minute time control, under this time pressure even Stockfish makes simple errors, therefore even an out of book score of let's say +0.8 can easily turn into into a lost game, because the time is not enough to find to correct continuation of a position which was in most cases calculated with a time control around 2-4 minutes.

Therefore you can find many positions where Stockfish makes simple errors and replay the lines which lead to those positions.

Second point, still not the important one:

Not only .ctg can learn of played games, also Cerebellum can. It is not public available yet, so this is debatable. But in theory, of course, you would have to feed this 1264 games into Cerebellum too, before playing the match. Of course you can't do this for now , I see this point.

Third point, the important one, I think.

After you won the match, and Cerebellum was updated with (calculated) the lost lines, you can't repeat your success without having access to the updated Cerebellum library. So the library is the tool you need, without pre-access to it it would most likely go the opposite way, because .ctg is statistics, and Cerebellum is calculated.

Thomas

[velmarin]

Such is the idea, how you drive at the end of your dissertation.
You can see that I've already said, it depends on how your tool manipulates the games and future chess problems, that only you knows it.
Your project would be to Playchess players or InfinityChess, if your tool is simple and efficient can succeed, in my opinion.

Somebody can expand this BrainFish vs BrainFish, I have hundreds of thousands of positions in suites from scratch.
In years past I spent to prepare these books ctg, then had tournaments in various forums, and I learned my tricks.
Also understand that your book is light, you want much luck.
Better, I think that the project will have that luck and his prize.

[Thomas Zipproth]

This project started about 10 years ago, where we had a complete Chess GUI (still not released) and I had the idea to integrate an analysis tool for Grandmaster opening preparation and Correspondence chess playing.

In the following years I played around with it a bit, using Rybka232a, Rybka3, and Houdini1,2,3,4, but I didn't had the time or resources to release something. Then I saw that with Stockfish, except the advantage that it is free, it started to work really good. That means, most positional problems could be solved, not only the tactical ones. At some point I programmed a proxy to play with the library connecting it to an uci engine.

Then I had the idea, before never releasing something, let's create a Stockfish Version which can directly use that library for playing, release it to the public and see how it behaves. I got very useful feedback so far, and it's fun too. And of course it may be some promotion for the full library, which may arrive finally at some point in the future.

I think one of the key points of the library is, that it can totally change it's behaviour when adding only some positions, because this nodes can connect already existing nodes, making new transistions. Also a new endgame node can change what is played at the beginning. That makes it so interesting to analyze positions or openings with it, which always was the main focus of the library.

[jefk]

[quote="Thomas Zipproth"]
Then I had the idea, before never releasing something, let's create a Stockfish Version [/quote]

well as already posted earlier, a certain arshah
system tries the same; but it's depending on stockfish;
which evals are not perfect. if you want to make
a nice book editor fine with me, but the brainfish
idea is confusing for some people and thus
not such a good idea; my recommendations are
make/keep it polyglot compatible, make a
nice book editor, and contribute to the computer
chess community in general; in the end chess
is a draw anyway.

[jefk]

[quote="jefk"] in general; in the end chess
is a draw anyway.[/quote]

which does not mean book moves are not important;
there are some sharp lines eg in Sicilian where you
have to play the best book move, otherwise you can lose.
Haven't checked brainfish for such sharp lines, but
it may very well be that not all the theory lines as
eg in poisoned pawn of the Najdorf are put in the
Bf/cerebel book in the best way; just a hunch.

anyway like i suggested i hope Thomas will not
remove this polyglot 2.03 compatibility, for the
rest i presume i was clear; not intended as
negativity, but i think we should compare (not run
tournaments) with apples (books) and oranges
(engines) if you see what i mean; for the
rest it's a free world isn't it; if i make a book
for stockfish/sugar and call it Supersugarfish
and try to sell it, that would be legal i suppose