Author: Miguel A. Ballicora
Date: 09:27:34 09/24/01
Go up one level in this thread
On September 24, 2001 at 05:37:11, Gian-Carlo Pascutto wrote: >On September 23, 2001 at 13:24:56, José Carlos wrote: > >> Of course, there're zillions of ways to handle this smartly. > >I handle this by searching on after a mate is found >until the nominal search depth = found mate depth. > >This way (except for nullmove madness) we are sure there >can be no shorter mate and we can still stop the search >early. > >I personally think that, even though it may not increase >the actual strength of the engine, this is important. But it increases the noise of the test. Some changes may make the engine find one or the other first. I agree that if there is a mate in 10 vs mate in 20 we can start considering the first as the only best move, but this case apparently is mate in 9 vs mate in 10. To make matters worse, mate in 10 is forced by checks and the first move is a capture and a check. Being a capture, there is no problem with any kind of bug mentioned before regarding not finding the shortest mate. The position is transformed so it eliminates any problem with repetitions and the like. A move that returns a mate score AND is a capture is always good. Besides, a sequence of checks avoid any kind of bugs with nullmove and any other weird problem. I will take the checks anyday if I have to wait only one more move. It is forced. Suppose that my engine finds Rxg2 in 10 seconds and changes to Bb7 in 20 seconds. Suppose that I make a change and now finds Rxg2 in 2 seconds and changes to Bb7 in 40 seconds. Is the change good for this particular position? I think so!! but if this position is included with only "bm Bb7" the test will tell me that it is a bad change! Regards, Miguel >One common example is programs seeing tablebase mates and >then giving away pieces for a mate in 50 while there is a mate >in 6. > >Either way the program wins, but I hate it if the latter >happens. > >-- >GCP
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.