Author: KarinsDad
Date: 09:43:26 09/28/99
Go up one level in this thread
On September 27, 1999 at 07:41:39, Lin Harper wrote: > Fritz 5.32 has, I believe, a 'book learning function'. This means > that it will avoid lines that it does'nt get good results from. > Presumably this means that if any of us were to purchase a new copy > of Fritz off the shelf, it would'nt play as strongly as SSDF's Fritz, > because there's has played a lot of games, and therefore plays stronger. > Unless I'm mistaken, this can only mean that SSDF's evaluation of > Fritz (and a few others) is rather meaningless. This is like saying that a new car that the manufacturer asks you to break in and keep under 55 miles per hour is not as good as a 6 month old exact same make and model car that is already broken in. Just because a program has a learning function to improve it's book does not mean that the engine is any stronger or weaker. So, if you were using Fritz for analysis, the version the SSDF is using would be the same as the version you are using (assuming identical hardware and settings). What a learning program can do is avoid certain lines. However, that does not invalidate the tests. In fact, the exact opposite occurs. The more games that are played, the more the tests validate the effectiveness (or ineffectiveness) of a learning program (in other words, the more games played, the closer you will get to a more accurate approximation of the rating). A program is a summation of it's parts, not just the strength of it's engine. So, the results become "meaningless" only when the person reviewing the results does not take in the overall picture and makes an implication that the results are equivalent to engine strength (which is a false premise). KarinsDad :)
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.