Author: bob o
Date: 11:48:01 03/25/02
Go up one level in this thread
>I just did a quick test to see if there is any gain through book learning. I
>loaded One computer with Fritz 7 and let it learn from the 3 databases I have
>(more than 1700 games played by Fritz 7. In the other computer (both AMD 1.4G)
>I cleared the book learning in Fritz 7 and played 100 games at G/1 minute (for
>quick results of course). The final score: Fritz without previous book
>learning won by 52-48.
1. Unless the final score was ~75-25 or more then there is not a significant
difference with just 100 games (I don't have my college stats book right
here)...
2. Such a short time control may introduce other errors that are unrelated to
the book learning (ie losing a won endgame due to time pressure)
3. Even without that type of error, there is always some small amount of random
variation that can't be controlled
4. I think it may be more interesting to find a line where there was book
learning from your databases. Then run a match where every game uses this line.
Even more interesting would be several long series on the same line. Perhaps
book learning would show up if the score of the matches turned out to be
increasing, such as:
Learn-No learn
Match 1 48-52
Match 2 51-49
Match 3 55-45
Match 4 57-43
Match 5 60-40
Bob
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.