Author: Dan Homan
Date: 04:54:41 09/16/99
Go up one level in this thread
On September 15, 1999 at 22:56:34, Robert Hyatt wrote: >On September 15, 1999 at 17:12:16, Dan Homan wrote: > >>I was reading the crafty source again the other day and noticed that >>Bob has a special function to improve the move ordering at the root >>of the search. >> >>I really didn't feel like writing such a function last night, but I >>thought instead to use the values returned by the search itself to >>improve the move ordering at the root. I know that I only get an >>accurate value for the best move, but I thought that my fail-soft >>search might return useful numbers for the other moves as well.... >> >>Implementing this was pretty quick and easy: there were a couple >>of pit falls, but the total changes were about 5 lines of code. >>Previously I simply used the same move ordering at the root that I >>use at all other nodes. >> >>The improvement was amazing! I got a full ply in many positions and >>about a half-ply in many more. It improved my solution times on WAC >>noticably and seems even better in quiet positions. >> >>I know that my solution was a quick kludge, so I am wondering what >>other people do for move ordering at the root of the search. >> >> - Dan > > >Are you sure you didn't introduce a bug? Root ordering is only important >in positions where you change your mind a lot. If you lock on to one move >and stick with it, it won't do anything to improve the ordering... I noticed (before the change) that about half of my search time (in some positions quite a bit more) was spent on moves other than the hash (best) move. I thought this was high and was looking at crafty for hints. That's when I noticed the special ordering at the root of the search, so I tried this quick solution of using the fail-soft values of the search to sort the list at the root. I don't think I introduced a bug, but I may well have covered-up a bug - I am going to look into this further, both to test how large the improvement really is and to see if I had any bugs that were limiting my performance before. > >a ply sounds really odd as a gain... I thought about this and realized that it is possible that the speed up was 'only' 10-20%, but because I count a ply as searched when it is completed and a half-ply as searched when the first move is completed, that I noticed many positions that seemed to get an extra half-ply or ply. - Dan
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.