Author: Toni
Date: 10:27:26 01/11/04
Go up one level in this thread
On January 10, 2004 at 22:38:59, Mike Siler wrote: >On January 10, 2004 at 20:44:33, Jaime Benito de Valle Ruiz wrote: > >>On January 10, 2004 at 19:09:34, Toni wrote: >> >>>Hi all, >>> >>>I've written the move generation routines for my engine. I'm now debugging them. >>>I've tested some positions in which there are castlings, en-passant captures and >>>promotions, and everything seems OK, but when computing perft sometimes it gives >>>wrong values at depth 4 or higher. The question is how to debug that, since the >>>number of moves involved is enormous. >>> >>>regards >>> >>>fermath >> >>Unless you want to create an automated way of checking this, when running a, >>let's say perft 4, and make sure your program displays the perft-1 (3) nodes >>that you get for every move, and compare them with the values of other programs. >>Pick one of the moves that has a wrong number of nodes, and make a perft 3 from >>that position. Again, you'll have a list of perft-1 (2) for each move, and at >>least one will be wrong. Repeat the process until you reach perft 1, and then >>you'll be able to see what move is illegal, and you'll be able to track the bug. >>I cannot remember which one, but there is a free program that has this "perft" >>and shows the node-count for every move rather than just giving you the total. >>Search the archives. >>Good luck, >> >> Jaime > >Albert Bertilsson's program "Sharper" is what you should use. The command >"divide x" will do a perft x and show you the number of positions resulting from >each root move. Also, Sharper is really fast, which is nice when you realize you >have a problem with perft 9 :) > >It's available here: >http://www.albert.nu/default.asp?sub=programs/default.asp?sub=dperft/main.htm > >Michael I was trying to do this, but I didn't know about the existence of programs that give the number of positions for each root move. This will be a great help, I guess Thanks
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.