09 May 2010

CCRL Discussion Board

While I'm on the topic of Label CCRL, I should mention the CCRL forum: CCRL Public. There isn't a lot of activity for chess960 -- 'Search found 13 matches: chess960' and 'Search found 63 matches: frc' -- but there are some interesting threads.

One thread from 2008, FRC Openings Statistics, complements a recent post, Four Weak Pawns (SP115), where I wrote about the value of the CCRL games in real play.

After some practical experience with CCRL games, I'm less enthusiastic about their value to chess960 opening knowledge. The engines don't play with a book (meaning they don't 'learn') and they have no notion of what it means to follow a strategic opening plan. In my own games I have often found an initial move that none of the machines had played and that seemed like a reasonable move to me. After playing my own moves a few times and suffering no unexpected drawbacks, I stopped looking at the CCRL collection for ideas.

The CCRL thread discusses practical problems of engine vs. engine testing. For example,

I don't think you can get meaningful statistics by repeating the same starting positions with the same engines 100 times. Because there is a large chance of repeating the whole game or the beginning of the game. The engine you are using has particular understanding of particular positions. Even if there is sometimes variation in the chosen moves, the results are not statistically reliable to say something about the position itself. All you can say is that in [a certain engine's] understanding the position is usually won by White.

It's a good point. See the thread for more good points.

No comments: