NEW Computer Chess Idea! middlegame books

[mainsworthy2]

Sorry Im not being clear here, hope I can get the point over here

Example
SonOfMainsworthy vs Rybka 0:1

Game analysed, now midgame postion where Rybka tipped the ballance, placed in middlegame book, then a responce to the position added.

Now with book only 200bytes long would hit every time(100%) against Rybka(asuming Rybka played the same moves against SonOfMainsworthy if Rybka playes a random type move then Rybka is a bad engine example)

You would then have the same game move being played upto midgame book position reached, and the resonse then would be played from the book, now the game from this point on would be different.

you build a midgame book like this, for each engine you play, so the books would get bigger and does not get its moves from opening books

but position learning sound like this may have already been done I dont know.

[Matthias Gemuh]

Sorry Im not being clear here, hope I can get the point over here

Example
SonOfMainsworthy vs Rybka 0:1

Game analysed, now midgame postion where Rybka tipped the ballance, placed in middlegame book, then a responce to the position added.

Now with book only 200bytes long would hit every time(100%) against Rybka(asuming Rybka played the same moves against SonOfMainsworthy if Rybka playes a random type move then Rybka is a bad engine example)

You would then have the same game move being played upto midgame book position reached, and the resonse then would be played from the book, now the game from this point on would be different.

you build a midgame book like this, for each engine you play, so the books would get bigger and does not get its moves from opening books

but position learning sound like this may have already been done I dont know.

This idea is horribly worthless for very many reasons.
Why don't you see it ?

Matthias.

[mainsworthy2]

Mat , I bow to your better understanding.

[krazyken]

asuming Rybka played the same moves against SonOfMainsworthy

This assumption is bad, for just about any chess engine. Use your favorite chess database (scid works well for this), load in a million or more games (no duplicate games), start at the opening position and browse the move tree, always pick the most popular move and see how many moves you get before there is only one matching game.

[Dann Corbit]

Sorry Im not being clear here, hope I can get the point over here

Example
SonOfMainsworthy vs Rybka 0:1

Game analysed, now midgame postion where Rybka tipped the ballance, placed in middlegame book, then a responce to the position added.

Now with book only 200bytes long would hit every time(100%) against Rybka(asuming Rybka played the same moves against SonOfMainsworthy if Rybka playes a random type move then Rybka is a bad engine example)

You would then have the same game move being played upto midgame book position reached, and the resonse then would be played from the book, now the game from this point on would be different.

you build a midgame book like this, for each engine you play, so the books would get bigger and does not get its moves from opening books

but position learning sound like this may have already been done I dont know.

This idea is horribly worthless for very many reasons.
Why don't you see it ?

Matthias.

Against a predictable engine (one without randomness introduced) it could be very useful.

It also has usefulness to automatically extend the book in the most frequently played direction.

Of course, against a strong engine with some degree of randomness, it will be hard to get a great deal of value out of this approach. But even at that it would be interesting to see experimental data results to show the actual, measured value.

[Uri Blass]

Sorry Im not being clear here, hope I can get the point over here

Example
SonOfMainsworthy vs Rybka 0:1

Game analysed, now midgame postion where Rybka tipped the ballance, placed in middlegame book, then a responce to the position added.

Now with book only 200bytes long would hit every time(100%) against Rybka(asuming Rybka played the same moves against SonOfMainsworthy if Rybka playes a random type move then Rybka is a bad engine example)

You would then have the same game move being played upto midgame book position reached, and the resonse then would be played from the book, now the game from this point on would be different.

you build a midgame book like this, for each engine you play, so the books would get bigger and does not get its moves from opening books

but position learning sound like this may have already been done I dont know.

I think that it is not different than opening book.
opening books can have many moves in them and I suspect that people already use this way to build opening books so your idea is not new.

In one of my correspondence game I am black and I play move 32 and the position is still theory based on the opening book that I use(I can see that the position can continue to be theory even many moves later)

Uri

[Matthias Gemuh]

Against a predictable engine (one without randomness introduced) it could be very useful.

It also has usefulness to automatically extend the book in the most frequently played direction.

Of course, against a strong engine with some degree of randomness, it will be hard to get a great deal of value out of this approach. But even at that it would be interesting to see experimental data results to show the actual, measured value.

That type of midgame book will not work even against the engine for which it was made, if you use a different time control or different computer.

BTW, storing a whole bookline just to avoid a single midgame move is strange economy, especially when no suggested best move is available.

Matthias.

[Matthias Gemuh]

Sorry Im not being clear here, hope I can get the point over here

Example
SonOfMainsworthy vs Rybka 0:1

Game analysed, now midgame postion where Rybka tipped the ballance, placed in middlegame book, then a responce to the position added.

Now with book only 200bytes long would hit every time(100%) against Rybka(asuming Rybka played the same moves against SonOfMainsworthy if Rybka playes a random type move then Rybka is a bad engine example)

You would then have the same game move being played upto midgame book position reached, and the resonse then would be played from the book, now the game from this point on would be different.

you build a midgame book like this, for each engine you play, so the books would get bigger and does not get its moves from opening books

but position learning sound like this may have already been done I dont know.

I think that it is not different than opening book.
...
Uri

It is worse than an opening book because (among other things) it has no suggested best moves.
If a particular line is played till midgame in such a book, that does not mean that the weaker engine played optimal moves.

Matthias.

[jhaglund]

I've already thought of how to do this. I am not capable of coding it.

I mentioned it to Robert Hyatt months ago using Crafty as the "brain" ...

I'm full of ideas.

http://www.talkchess.com/forum/viewtopi ... ew=threads

[mainsworthy2]

It would be nice to see it in crafty Joshua.