Suggestions for new CCT rules

[Harvey Williamson]

You've got a point I did not though of. Of course using Rybka vs rybka games to generate a deep book for an engine other than Rybka seems un-legitimate.

Therefore generating a book from GM v GM games is also unfair.

[mathmoi]

i dont want that specially cct plays without books, i wanted only propose that there could be one other tourney without books.

I'm curious, but I don't think many author would be interested to participate in such tournament. I know I would not because I think that the opening book is an integral part of my engine.

(Note that my engine generate it's own book from a PGN of ranked players and engine vs engine games. It still evaluate each position with a small search to decide if it stays in the book or not.)

[garybelton]

However I don't know how we could prevent one while still allowing the other.

That's why my belief is the only way to cure it is to limit the book depth, then it doesn't matter so much where the moves originated from.

Another post mentioned limit the size of book in MB on disk. But that would need to be done by book format. You can create a maxmoves=unlimited book from a set of games CTG (200mb) Aquarium (100mb) Polyglot (10mb) - same games make different sized books.

[mathmoi]

the book is part of the engine, you can't amputate it

Good input, but you know that many top opening book authors find the moves that go into these books using a combination of engines other than the engine that is playing with it? (Along with their chess skills). Like in the Sjeng-Hiarcs 50 move game in book, this was forged out of 1000's of Rybka v Rybka hours on the Playches server. So I don't really agree that the book is part of the engine, in actual fact you are playing moves found by other engines from this book.

Hi Garry,

You've got a point I did not though of. Of course using Rybka vs rybka games to generate a deep book for an engine other than Rybka seems un-legitimate.

What I would not want to see disappear is book created from, let's called them historic games. That is human and engine games that were not played for the sole purpose of creating a opening book. I also think that programmer should still be able to use book learning capabilities, even if this mean that they can beat a particular engine while still in the book.

However I don't know how we could prevent one while still allowing the other.

Therefore generating a book from GM v GM games is also unfair.

It's not my opinion. All the opening culture we currently have has at a moment or another in the past been found by highly ranked player. Now engines are starting to find new lines that highly ranked human will play.

So human and engines have always looked at what highly ranked player played to decide what to add to their personal repertoire and this is fine in my opinion.

What Garry was describing is an author using a stronger engine to generates his engine's book. Personally I only use 1) games that are public (enormous.pgn, games played on ICC and FICS, CCRL, etc.) and 2) games played by my own engine. So if I wanted I could run a really long match of MatMoi vs Rybka to create my book or a Rybka killer book. I could also use Rybka games avaible online, but I could not use Rybka to generate games to then import in my engine book.

I think that mimics what human had always done. You can study the public game of your opponent and copy their moves, but you can't ask him "Hey would you play a couple thousand game for me? I need to prepare for a match and I'd like to use your opening repertoire."

[Harvey Williamson]

Such a rule would be impossible to enforce. You can study the public games of your opponent if its an engine by watching games on Playchess, ICC, FICS, SSDF, CCRL, CEGT etc..

[garybelton]

The other point Harvey is that CB sysops (like you) have access to all the engine room games. I know this because a previous sysop kindly send me a subset of them and also put a lot of them online at some point. This is a great advantage.

[Harvey Williamson]

The other point Harvey is that CB sysops (like you) have access to all the engine room games. I know this because a previous sysop kindly send me a subset of them and also put a lot of them online at some point. This is a great advantage.

I do not have access to the Playchess DB any games I collect, I collect myself.

[mathmoi]

However I don't know how we could prevent one while still allowing the other.

That's why my belief is the only way to cure it is to limit the book depth, then it doesn't matter so much where the moves originated from.

Here is what could happen if you do so.

In the first game of a match engine A beats engine B. Later in the match for some reason the same position out of the book happens again. Engine B, because it is capable of learning know that it should not do a certain move, because last time against this opponent it lost the game, however it is forced to ignore that fact and play without the learnt information.

I know it's a stretch, engine B should I avoided to get in the same position in the first place. But what I'm trying to expose is that this rule will penalize the learning engine in favor of other.

You want to limit the practice of using the moves from another engine and doing so you remove some of the incentive to create engine that can generate and learn their book.

[garybelton]

I do not have access to the Playchess DB

As CB sysop why do you not have this access?

[Harvey Williamson]

I do not have access to the Playchess DB

As CB sysop why do you not have this access?

I only know of 1 sys op with access and they are a paid member of CB staff.