Author: Robert Hyatt
Date: 07:59:27 10/14/03
Go up one level in this thread
On October 14, 2003 at 10:26:20, Uri Blass wrote: >On October 14, 2003 at 10:06:41, Robert Hyatt wrote: > >>On October 14, 2003 at 08:42:05, Tord Romstad wrote: >> >>>On October 14, 2003 at 07:17:27, Omid David Tabibi wrote: >>> >>>>Actually it seems that it is safe to ignore all capturing checks altogether in >>>>gen_checks(). If the capture is non-losing, it will already be generated, and if >>>>it is a losing capture, the odds that it turns out to be a good move aren't that >>>>high (assuming that SEE hasn't mistakenly deemed the move as losing capture, >>>>while in fact it is a winning capture since one of the defenders is pinned...) >>> >>>I think it is a good idea to use the SEE for checking moves, too. Only search >>>checks which do not lose material. >>> >>>Tord >> >> >>Then you admit defeat before you start. See the simple draw by stalemate >>idea where you have a king and rook, and your king can't move. You just >>make unsafe check after unsafe check because the rook can't be taken. >> >>Safe checks are probably better than no checks, assuming you believe you >>have to do checks in the q-search. But, as I have said so many times, if >>your q-search is going to have so many holes, find a way to avoid putting >>such a heavy responsibility on the q-search in the first place. > >why do you assume that it is heavy responsibility? The function of the q-search is to take a leaf position, and eliminate any dynamic tactics so that the evaluation can produce a number that is reasonably accurate. Captures are a small part of potential tactics. What about checks? Skewers? double attacks? Forks? mate threats? Threats to promote? Overloaded pieces? The q-search only considers a tiny part of that set of tactical themes. Yet we depend on it to give quiet positions to the evaluator. That _is_ a "heavy responsibility". It is where most programs make their tactical errors. IE the "horizon effect" is a fault of the q-search, not anything else. >The fact that you do not get everything from checks in the qsearch does not mean >that it is not productive. Of course not. But it does mean that the q-search has plenty of errors left. Checks just make it bigger. Which gives even more opportunity to produce bogus scores... > >The test is simple >If the program plays better then it is a good idea. Agreed... > >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.