Author: Howard Exner
Date: 21:49:39 03/18/98
Go up one level in this thread
On March 18, 1998 at 14:08:05, Amir Ban wrote: >Of the 40 Fritz5-Rebel8 games, a staggering 16 are duplicates (2 groups >of 7, 1 of 3, and 2 of 2, there would be more if not for random >fluctuations out of book). They contain 10 wins for Fritz, and 6 draws. > >If you omit the duplicates, instead of 31.5-8.5 (78.75%), the result is >18.5-5.5 (77.08%). Hardly a difference worth mentioning. The SSDF say >that doubles are not statistically significant. This result support the >argument. From observing the SSDF formula it appears that each gain of 1 % is worth 7 rating points (can anyone confirm this for sure?). So in this small sample Fritz gains about 12 rating points (for this match only, not it's rating based on all games played). > >The number of doubles are peculiar to Rebel. I didn't see any duplicates >in the match against Junior 4.0, though it doesn't have any learning >whatever. The reason is that it plays with much variation in the opening >book. Getting it to repeat a line is way against the odds. The reason >Rebel has so many doubles is that with its tournament book it has a >rather small repertiore that makes repetition likely. I've seen doubles before but what is different about Fritz's opening programming that makes this so pronounced when playing Rebel 8? Rebel 8 has played other learners but has never experienced a match with over 40% of the games being doubles. I've always understood learners to be what I witnessed from the Rebel 8 vs MCPro 5 SSDF match where when MCPro 5 lost or drew a game it replayed the line again but with an added change to the line (ie: it didn't just dump the line but tried to fix it). > >By the way, as the other one at the receiving end of this F5 onslaught, >I can say this: I wish the result was better, but I have no problem >accepting it as true. It's hard enough to keep up tactically with Fritz >on equal hardware. If I am invited to a game against it with 3:1 CPU >odds, I would rather not show up, and send a resignation by mail. Having Junior play a wide variety of lines is also a good consumer feature. I remember when opening books were so narrow that I was forced to manually enter moves in for my computer opponent by using opening book manuals. That was a nuisence. > >Ed has obviously invested much in the opening book, and the tournament >book it seems was carefully handpicked. I guess Ed was not aware of the >possible implication of playing a small repetoire. It leaves him open to >victimization by learners. In the end, I don't see that victimization >really took place, but anyway no one is forcing him to continue this >practice. > >Ed, my own conclusion from this match is that Rebel was hopelessly >outgunned tactically, and that the duplicates were a curious feature of >it that didn't really change the outcome. If you believe otherwise, I >think you are deluding yourself. > >I posted a suggestion about what should be done about this recently. I >guess I was a bit unclear, so I'll do it in more detail soon.
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.