Author: Mike Hood
Date: 22:52:02 06/28/99
Go up one level in this thread
On June 28, 1999 at 12:51:42, Melvin S. Schwartz wrote: > >On June 28, 1999 at 01:24:49, Mike Hood wrote: > >>On June 27, 1999 at 22:06:48, Melvin S. Schwartz wrote: >> >>> >>>On June 27, 1999 at 15:24:33, Gustavo Pereira wrote: >>> >>>>I have seen many posts where they say Fritz cannot solve this or that problem. >>>>Many of the times it is because Fritz is (by default) a null mover. >>>> >>>>Now what is the reaction? >>>> >>>>Many people start saying that it is because Fritz is not a good program, and >>>>that you can't switch off the null move engine. >>>> >>>>How far from reality!!! >>>> >>>>Quoting from the (really meager) Fritz 5.32 manual >>>> >>>>'(Selectivity) Denotes the number of plies reduced by the null-move. A value of >>>>zero means that the null-move is switched off'. >>>> >>>>I guess that says it all. >>> >>>What effect would changing the default setting to zero have on Fritz's playing >>>strength? >>> >>>Mel >> >>This is a theoretical question, probably better left to a chess programmer, but >>I'll be the first fool to rush in. >> >>Many months ago I compared Fritz's performance with different Selectivity >>settings. The higher the value of Selectivity (6 is maximum), the deeper >>Fritz searches, ie the more plies it examines in the same time. At Selectivity 0 >>Fritz doesn't search as deep, presumably because it examines lines it would have >>ignored otherwise. However, the word "Selectivity" seems to be misleading, >>because even at Selectivity 0 Fritz seems to be doing some "selection" of lines, >>it seems to be too fast to be doing a Brute Force search. >> >>To put it succintly: the lower the Selectivity value, the less deeply Fritz >>searches, but the more thoroughly it analyzes. This means that by setting the >>Selectivity to 0, Fritz might avoid blunders like in the cute little chess >>problem Kurt quoted, but Fritz could make other errors by not searching deep >>enough. >> >>My guess (here is where I need backup from chess programmers) is that blunders >>through null move errors are relatively rare. The advantages of deeper searches >>far outweigh these occasional problems. >> >>Speaking practically... Fritz has seven different Selectivity values, from 0 to >>6, and Chessbase has made 2 the default. I assume that this value has been >>proven to produce the optimum playing strength, either experimentally or >>algorithmically. I'm not sure if a higher Selectivity value brings advantages. >>Maybe someone can present a position in which "Fritz Sel 6" finds the best move >>in 10 seconds, whereas "Fritz Sel 2" needs 5 minutes. I'd be glad to hear about >>it. > >Hello Mike, > >Actually I knew all that believe it or not. The reason I put up the question was >to hear if someone actually had found Selectivity at zero producing a higher >playing strenght, which would have started a whole new debate. The other reason >I brought this up is because it has been said by Shep that his modified settings >for CM5000K result in superior play. Therefore, it is not unreasonable that a >possibility exists where the default setting is not optimum. > >Regards, >Mel> >>Mike After reading your message I decided to run a little test the only way I know how. I ran the Fritz engine against itself in an engine-engine match with Selectivity parameters of 2 and 0 respectively. After 12 matches Fritz-Sel-2 had won 8, lost 2 and drawn 2. That's admittedly only a small sample, but it confirms my suspicion that the default setting makes the playing strength stronger. Maybe we could get a more interesting result if we played both versions of Fritz against a stong non-null-mover program. That's beyond my testing capabilities though. Mike P.S. If you want to try this yourself, I had to resort to tricks to play Fritz against itself with different parameters. I had to make a copy of the engine file, Fritz5-32.eng, then edit the copy to fool the main program into thinking it was a different engine.
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.