Author: blass uri
Date: 11:23:38 08/03/99
Go up one level in this thread
On August 03, 1999 at 10:27:49, Robert Hyatt wrote: >On August 03, 1999 at 09:49:44, blass uri wrote: > >> >>On August 03, 1999 at 09:14:30, Robert Hyatt wrote: >> >>>On August 03, 1999 at 05:25:52, Ed Schröder wrote: >>> >>>>>Posted by leonid on August 02, 1999 at 21:23:37: >>>>> >>>>>>IMHO low-brain fast-searches like DB vs Kasparov have proved it is better to >>>>>>forget about trouble makers and exceptions and just go for the brute force >>>>>>approach. Fast and dumb rules. Forget about exceptions they are waste of >>>>>>time. >>>>>>You spend all clock cycles and programmer time on worrying about >>>>>>exceptions and then you are full of bugs. >>>>>> >>>>>>Ciao >>>>>> >>>>>>Mark >>>>>> >>>>>> >>>>>>> >>>>>>>And because of today's fast computers the exceptions fade away as for >>>>>>>example the Cray Blitz position is seen by Rebel in 0.5 second. >>>>>>> >>>>>>>Ed Schroder >>>>> >>>>>I really agree with what was said obove. Now on very quick computers Rebel >>>>>10 can see by "brute force" 6 plys ahead in just one or two seconds. Some >>>>>less superficial revision of the moves but with "fixed horizon" can lead up >>>>>to 10 or even 12 plys deep. This way of searching the move is best >>>>>that some other method that care too much about exceptions. Exceptions >>>>>that take that much space to care about and can produce anyway very >>>>>suspicious result. >>>>> >>>>>Leonid. >>>> >>>>I do not agree with was has been said above except what has been said >>>>by myself of course :-) >>>> >>>>If you have a commercial program and playing a 40/2:00 game for instance >>>>you can not afford to think 6 minutes (or worse) on a simple recapture as >>>>people are going to laugh on the stupidness of the silicon. >>>> >>>>So you are forced to come up with some intelligent software that handles >>>>forced moves. This means you are going to have to deal with all the >>>>exceptions. No choice. >>>> >>>>Ed Schroder >>> >>> >>>That is debatable... I think your reasoning is a dead match for the reasons >>>that Slate/Atkin used for their famous "that was easy" idea in chess 4.x... >>>they didn't like sitting for N minutes on an obvious recapture. Many of us >>>didn't want to look silly like that. And often (or probably all of the time >>>in fact) the fix was actually worse than the "problem". But we didn't realize >>>this until we got burned once... >> >>The fact that you lost one game because of this is not a proof that the fix is >>worse than the problem because it is hard to tell how many games you lost >>because of the problem. >> >>It is possible that saving time help you to find slightly better move later in >>the game and it is not easy to know if the slightly better move gives you a >>better result or does not give you a better result. >> >>Uri > >that was one game that went from potential draw to dead lost due to our 'that >was easy' approach. I initially did this same approach in Crafty, but saw it >lose enough games on ICC that I greatly restricted it to the point where it is >now. And I haven't seen the current approach lost any games, although the >possibility is admittedly still present... > >But it definitely lost plenty of games on ICC before I got rid of it. I think that maybe it lost more games on ICC at bullet and blitz and the question is what is the probability that a move that crafty considers as forced is wrong and there is not a big drop in the evaluation after using 1/3 of the regular time. In the case of the move of cray blitz Crafty can suspect that the move is wrong even before it finds the right move because there is a big drop in the evaluation. I think if it happens in less than one out of 200 games at standard time control then the benefit of saving time is bigger than the demage. Uri
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.