Author: Robert Hyatt
Date: 20:03:03 11/21/00
Go up one level in this thread
On November 21, 2000 at 16:02:38, Mike S. wrote: >On November 21, 2000 at 14:48:25, Robert Hyatt wrote: > >>On November 21, 2000 at 14:32:51, ERIQ wrote: >> >>>or is either better ?? your thoughts. >>> >>>I was thinking about buying yet another program for analyses of games. >>>(...) >> >>I don't like back to front. It requires that stuff from the end of the >>game be back-filled through the hash table. Here are the two arguments: >> >>1. back to front is better... because you can use stuff happening at the end >>of the game to influence scores and moves earlier in the game. >> >>2. front to back is better because it gives a true reflection of what the >>computer would have seen had _it_ been playing that game directly. >> >>(...) > >From my experience with Hiarcs 7.32, I can tell that 1.back to front >impressively enhances the analysis, especially if you step through certain >variations back and forth manually (I think H7.32 was the first one with the >backward feature of this kind; Fritz 5.32 had it with the special function only >AFAIK). For example, the program kind of carries mates upwards through the game, >which results in a mate announcement much earlier than possible in a front to >back analysis, given the same time. Of course, the same logic applies to any >kind of continuation, as long a the point of a move or idea is contained in the >notation (!) the program calculates on. The problem is that this info is carried backward thru the hash table. If it gets overwritten, then suddenly the program is ignorant about what it thought previously, and the scores change significantly, although the game itself has not changed... > >Therefore, it's clear to me that back to front is not useful for testing the >program (instead of the game's moves). But testing requires to know the correct, >or best, answers already. On the other hand, I can imagine that for the analysis >of test games, it is good to begin with 1. back to front as a first step to find >the "peak situations" of the game, which are the important positions to start >the tests. > >But for the analysis, if the point has NOT been played or included in a >variation ahead, the "normal" search which is applied anyway for each single >move (too) is utilized, i.e. by the Fritz GUI's analysis feature. Btw. it >produces variations during this too. > >In short words, my recommendations: > >Game analysis: back-front >Program testing: front-back > > >Regards, >M.Scheidl
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.