Author: Robert Hyatt
Date: 05:19:56 07/20/00
Go up one level in this thread
On July 20, 2000 at 02:37:07, blass uri wrote: >On July 19, 2000 at 23:42:46, Robert Hyatt wrote: > >>On July 18, 2000 at 14:29:53, blass uri wrote: >> >>>On July 18, 2000 at 14:27:58, stuart taylor wrote: >>> >>>>On July 18, 2000 at 12:17:57, blass uri wrote: >>>> >>>>>On July 18, 2000 at 12:10:57, stuart taylor wrote: >>>>> >>>>>>Let me clarify. Of course computers always look for the best move in attack, and >>>>>>also play the move they evaluate as as best. But when they see the evaluation >>>>>>going down, they will look perhaps a whole extra ply deeper in search of finding >>>>>>a way to bring the evaluation back up again. >>>>>> But if they see their evaluation going up, they do that move, sometimes very >>>>>>quickly, atleast they don't grind away labouriously trying to find something >>>>>>even better, even if the first "good move" is not all that good relatively. >>>>> >>>>>My knowledge is that Junior almost always finish the iteration of moves. >>>>> >>>>>If it see the evaluation goes up it will usually not play without checking all >>>>>the other moves in the same iteration. >>>>> >>>>>Uri >>>> >>>>Hiarcs too, usually. But if it is losing, it just goes on and on thinking, for >>>>very long periods, letting its clock tick till quite near the end. >>>> If that would help it salvage the position, it would be time well spent. >>> >>>Hiarcs is the only program that I know that does it. >>> >>>Uri >> >>Crafty will do it also... > >I meant to the fact that hiarcs does not finish the iteration of moves when it >fails low but does another iteration. > That has _got_ to be wrong. Why on earth would anyone write a search that would fail low, and then start a new iteration (that it might not have a chance of searching even the first move on) without first checking all of the remaining moves at this iteration to see if one will bring the score back up? I can't imagine any reason to do that... valid or invalid... >I claimed that Junior usually finish the iteration of moves before playing. > >stuart replied >Hiarcs too but if it is losing it just goes on thinking... > >He meant that finising the iteration of moves is not enough after failing low >and it does another iteration. > >I replied that hiarcs is the only program that I know that does it. > >My opinion is that the idea of doing another iteration after failing low may be >a good idea only when you have many moves to the time control. I would say it is senseless, period, if you don't first finish the _current_ iteration. Starting a new iteration with no clue about the best move is a good way to burn lots of cpu time and get nothing back. If Hiarcs does that, I'd bet it is a bug. > >If you have 110 minutes for moves 15-40 and you fail low then using 30 minutes >for one move because of doing another iteration is not a disaster and may be >productive but later in the game it is not a good idea. > >Uri This depends. If you have 45 minutes to make 5 moves, and you use 30 minutes to avoid losing a piece, it makes sense. As the extra time is of no use if you are essentially lost...
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.