2 Moves Openings Book

[Lyudmil Tsvetkov]

Hi.

Unbelievable as it might seem, I have prepared a very short 2 moves openings book for engine testing. The desire to do so was born out of the necessity to have an extremely short and perfectly equal book so that engines could play themselves a larger share of the game with no bias in the starting positions.

Some facts about the book:

- it contains 913 unique starting positions
- all of the lines are perfectly equal, i.e. there are no positions where one of the sides is lost; I guess the advantage in all cases is in the range of 20-30 centipawns at most, it is true that some engines might see bigger advantages, but I would definitely state that if there are any lost positions at all, there are below 1%
- for the sake of equal play, I have included lines like 1.d3 d5 2.c3 e5, which is still not lost for white, but excluded lines like 1. g4 or 1.b4, where white is already exposed beyond repair
- there are a lot of lines that should be rare or extremely rare, but still equal

The book is targeted at testing top engines, where already every single mistake counts, and so using a suboptimal book with losing variations very much distorts unbiassed testing. But of course, you can use with whatever engine.

Why the book is 2 moves short and not 1 move or 3 moves? Well, a 1 move book would not provide the sufficient variety of openings for proper testing, while a 3 moves book already unnecessarily takes another move of free choice for the engines.

Why I did not try to create a 2 moves book by using grandmaster games or games of lower rated players? Well, I suppose that openings in GM play would not provide sufficient variety, while in a larger sample of games of lower rated players there would surely be many suboptimal lines like 1. g4 and 1.b4 for example, so that you would still have to do some manual corrections.

The book is available for donwload here:
(see below)

Now, although I have written the book completely by myself, I must say that 90% of the credit goes to Mr. Adam Hair, our beloved mod, who made the impossible, sacrificing a lot of his personal time even during the weekend to convert the original doc file, delete any duplicates, and even add 3 other book formats to the pgn file.

The book, thanks to Adam, is available in 4 formats in all in the compressed file: pgn, Arena/abk, epd and bin, but I guess that testers could convert further to ctg, Shredder, Chess GUI, etc. formats themselves, or maybe someone could helpfully provide a download of those formats.

I expect that engine testing with such a short book will exhibit some peculiarities to the standard rating lists, as engines in general are not very well prepared to play the opening themselves, as simply their evaluation and search parameters are tuned starting with a stage of the game somewhat further the chess tree.

Would be interested to see if there will be some testers using this book, and of course, to pointing out how stupid, unequal, biassed, etc. the lines are, although I do not think I will be able to make any changes to the existing lines.

Any comments very much appreciated.

And many thanks again to Adam. (vote the mods for a 3rd term )

Best, Lyudmil

EDIT:
New link (with 1000 positions): http://www.mediafire.com/download/jskag ... LT_1000.7z

[lkaufman]

I very much like your idea. Maybe we will try using it. It would be nice if you could expand it to 1000 positions so that a complete test will run an even 2000 games. That would of course require slightly relaxing your requirements.

[Lyudmil Tsvetkov]

I very much like your idea. Maybe we will try using it. It would be nice if you could expand it to 1000 positions so that a complete test will run an even 2000 games. That would of course require slightly relaxing your requirements.

Hi Larry, thanks for posting.

I guess that for the time being it will stay like that, as I also depend on Adam for the conversion process, and he already did more than enough. But if there is interest, we will see.

In any case, I think that almost a 1000 positions is not that few, when you switch sides that makes 2000, a tourney with 3 participants would provide 6000 unique games, and when you run a gauntlet with 10 opponents, 20 000 unique games would be possible.

I guess in some of the lines the engines will give bigger advantage, but I am sure equality is still there.

[fauzi]

what format is the book ? (.ctg , .abk , .bin ) ?

[carldaman]

I very much like your idea. Maybe we will try using it. It would be nice if you could expand it to 1000 positions so that a complete test will run an even 2000 games. That would of course require slightly relaxing your requirements.

Hi Larry,

I think Don had proposed something similar at one point, an opening suite where the book ends after each side has moved 2 pawns (randomly) to the 3rd rank.

I think 784 unique positions would result, some more 'equal' than others
It should be named after Don.

Or it could be merged into what Lyudmil has.

Regards,
CL

[Adam Hair]

what format is the book ? (.ctg , .abk , .bin ) ?

pgn, epd, abk, bin

[michiguel]

I very much like your idea. Maybe we will try using it. It would be nice if you could expand it to 1000 positions so that a complete test will run an even 2000 games. That would of course require slightly relaxing your requirements.

Hi Larry, thanks for posting.

I guess that for the time being it will stay like that, as I also depend on Adam for the conversion process, and he already did more than enough. But if there is interest, we will see.

In any case, I think that almost a 1000 positions is not that few, when you switch sides that makes 2000, a tourney with 3 participants would provide 6000 unique games, and when you run a gauntlet with 10 opponents, 20 000 unique games would be possible.

I guess in some of the lines the engines will give bigger advantage, but I am sure equality is still there.

I think it may come a bit short. Most of the time you need more than 20k games and sometimes up to 80k games. I think that going to 3 moves will be useful, but I figure it could be very tedious.

Miguel

[Lyudmil Tsvetkov]

I very much like your idea. Maybe we will try using it. It would be nice if you could expand it to 1000 positions so that a complete test will run an even 2000 games. That would of course require slightly relaxing your requirements.

Hi Larry, thanks for posting.

I guess that for the time being it will stay like that, as I also depend on Adam for the conversion process, and he already did more than enough. But if there is interest, we will see.

In any case, I think that almost a 1000 positions is not that few, when you switch sides that makes 2000, a tourney with 3 participants would provide 6000 unique games, and when you run a gauntlet with 10 opponents, 20 000 unique games would be possible.

I guess in some of the lines the engines will give bigger advantage, but I am sure equality is still there.

I think it may come a bit short. Most of the time you need more than 20k games and sometimes up to 80k games. I think that going to 3 moves will be useful, but I figure it could be very tedious.

Miguel

It is true, but most ratings lists test with some 1000-2000 games. Ingo was 1250, similar CEGT, CCRL, and Stefan runs gauntlets.
Besides, on the 3rd move quite some tactics appear and I would myself shrink from the responsibility to insert tactical lines. Let the engines play themselves! For me, books are a remnant from the very distant past, when you had to build a book so that your engine would play at least reasonably well in the beginning phase. Now things have already changed beyond recognition, and I am seeing unexpected and probably very strong engine moves very early on (on a par with much misunderstanding). So, I would say, let the engines play themselves.

[wolfman]

Excellent work !
Question : do I miss the f4 openings ?

Eran

[fauzi]

Hello Lyudmil , Well done, interesting opening book for engine tests , if You don't mind , I can take the .abk book and add some extra lines with 2 moves (4ply) including the f4 openings that are missing to make the total of the opening lines to 1000 lines like Larry proposed , and also I can play a little with the priority of the moves to give each move the proper weight and probability for the .abk book (ofcourse if someone want to make a tournament using all the opening set then the priority of the moves will be useless , but still can be useful when someone want to test engines using this book but don't want to run the whole 1000 position, then the most playable lines will be played more often after I change the priority of the moves)
so do I have the permission to do that Lyudmil ?