From: Pauli Misikangas CS HELSINKI FI> Date: 22 aug 1998 Subject: Re: Sekita shogi On Fri, 21 Aug 1998, Jonathan Tisdall wrote: > >I'm not sure what is gained when the program drops the pawn on the same > >square like this several times. That's my point. Seems a waste to me. > >I get several free moves. I'd like to know the point of this if it isn't a > >programming flaw. > > This is in fact one of the things I was wondering about, along with its > general playing strength at higher levels. I have also experienced that it > will go into a kind of loop, repeating a time losing sequence. I have seen > it do the kind of thing you report by dropping a pawn to start, so that it > achieves absolutely nothing, simply loses a move. > > Still waiting to hear more. Well, I haven't tested Sekita Shogi on its higher playing levels, but I can make a quess why this happens. Earlier versions of my shogi program, Shocky, suffered from exactly same problem. I believe that this wasting of moves is caused by so called "Horizon effect" which is a common problem for all shogi/chess programs. Because the search depth (how far the program can see) is limited to some number n, the program might decide to make a move that looks good when searched to depth n, but turns out to be a catastrophe (or waste of time like in your examples) when searched one move deeper. In your pawn-pushing example, if the search depth is set to 3, the program sees the following moves: - computer moves/drops its pawn in front of your pawn - you capture it - computer captures your pawn with silver So the program thinks that it has gained some advantage because his silver is now one step further. In reality, you will drop a pawn in front of the silver and it must back off, but because the search depth is 3, the computer does not see this. Very likely the search depth used in Sekita is much higher than 3, but the problem is still the same. Also, programs do this kind of time waisting moves to delay some bigger loss. For example, if computer is about to lose a silver, it might sacrifice a couple of pawns because doing so pushes the capture of the silver beyond the horizon. Have you ever wondered why some programs start to do checks like a maniac when it's going to be mated soon? (if you don't know the answer, read this mail again :-) All the best, Pauli Misikangas ------------------------------------------------------------------ Pauli ? +358 9 3743257 (home), +358 9 70844271 (work) Misikangas \\|// Pauli.Misikangas cs Helsinki FI (o o) http://www.cs.helsinki.fi/~pmturune/ ----------oo0-(_)-0oo---------------------------------------------