Author: Don Dailey
Date: 22:35:03 01/14/98
Go up one level in this thread
Hi Bob, Do you consider fractional plys in your hash table implementation? Is a depth of 3 3/4 better than 3 1/2? - Don On January 14, 1998 at 14:25:41, Robert Hyatt wrote: >On January 14, 1998 at 08:49:27, Dan Homan wrote: > >>On January 13, 1998 at 14:43:16, Robert Hyatt wrote: >> >>>On January 13, 1998 at 09:08:49, Dan Homan wrote: >>>> >>>>p.s. My program does >>>> >>>>capture extensions (4/10 of a ply) >>> >>>dangerous one there. Ken Thompson used 1/2 ply in 1983, and probably >>>lost the 1983 World Championship as a result of doing this. He reported >>>it helped his WAC results quite a bit, but it costs about 1/2 to 1 ply. >>>He later discarded it totally. >>> >>> >>>>re-capture extensions (1 ply) >>>>check extensions (1 ply) >>>>pawn push extensions (1/2 ply or 1/3 ply) (endgame and late >>>>middle-game) >>>>bruce's mate ext. (1 ply) >>> >>>All reasonable. I am now using 3/4 ply for *everything* in Crafty, but >>>do allow multiple extensions to add together, with a limit of 1 ply. >>>This >>>actually improved the WAC results and seems to be running a good bit >>>faster >>>to reach the same depth... >> >>You said you allowed multiple extensions to add together with a limit of >>1 ply. I assume this means that you do not allow left over fractions to >>roll over to the next ply. > >your wording makes it hard to answer... but here goes: At every ply >along the way, depth gets reduced by 1 ply, as you would expect. At >every ply along the way the depth gets incremented by fractional values >that can not exceed one ply. So if ply 3 increments by 3/4, and ply >4 increments by 1, that 3/4 is *not* lost because it came from a >different ply. > >I only restrict a single ply to adding in no more than a one ply >extension... but it passes along everything it inherits from prior >plies. So a 3/4 extension at ply=1 gets carried along with no real >effect on the search until someone further along adds in at least 1/4 >more. Since all of my extensions are currently 3/4, the first 3/4 >doesn't >do much (this is a lie, more later) but the second 3/4 rolls this up to >1.5 >plies which extends every line below this by 1 ply. And the extra 1/2 >is >kept since it didn't come from the current ply, so that the next 3/4 ply >will also extend by one, as will the next (but not the 4th). > >I said the above is a lie because my initial search depth for a 1 ply >search is set to 1.75, so that the first 3/4 ply search extension does >something, as does the 2nd and 3rd. but the 4th won't, but then we >start >over again with the 5th, 6th and 7th extending but the 8th not. > > > >> >>I tried eliminating the capture extension, it reduces my search tree >>by a factor of 2 in most positions so I get about 1 ply deeper in the >>same time. I solve some of the WAC positions a ply later, but the >>speedup generalizes to all positions. I ran overnight with the new >>version on FICS, and it doesn't seem any worse (it even gained 20 >>rating points). >> >>I'll probably write some analysis tools like Bruce suggested to get >>a more concrete measure of the change, but I am happy with a smaller >>search tree. :) >> >> - 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.