Author: Ross Boyd
Date: 14:43:23 03/19/05
Go up one level in this thread
On March 19, 2005 at 06:26:28, Dieter Buerssner wrote: >On March 18, 2005 at 17:26:03, Ross Boyd wrote: > >>However, with TRACE I have become envious of other engine's PV lengths... (it >>must be a male thing, ha ha) > >:-) Hey, my engine just at the moment when I read this, spit out: > > 4523931 4.344 1.40 19. 49.Kd2 Ke8 50.Ke3 Kf7 51.Kf4 Kf8 52.Ke4 Ke8 > 53.Kd3 Kd7 54.Ke2 Ke8 55.Kd2 Kf7 56.Kc2 Ke8 > 57.f7+ Kxf7 58.Kc3 Ke8 59.a5 bxa5 60.Kc4 Kf7 > 61.Kc5 Ke8 62.Kd6 Kf7 63.Kc7 Ke8 64.Kc6 Kf7 > 65.Kb6 Ke8 66.Kxa6 Kf7 67.Ka7 Ke8 68.Kb7 Kf7 > 69.Kc7 Ke8 70.Kd6 Kf7 71.Kc5 Ke8 72.Kb5 Kf7 > 73.Kc4 Ke8 74.Kc5 Kf7 75.Kd6 Ke8 76.Kc6 a4 > 77.Kb6 Kf7 78.Ka6 a3 79.Kb5 Ke8 80.Kb4 Kf7 > 81.Kc3 Ke8 82.Kd2 Kf7 83.Ke2 Ke8 84.Kf3 Kf7 > 85.Kf4 Ke8 86.Kg5 Kf7 87.Kh6 Ke8 88.Kxh5 a2 > 89.Kg5 a1=Q 90.h5 HOLY SMOKES!!! I wouldn't dare spit out a PV of that length for fear of crashing the GUI. I know old Fritz GUI gets upset with long PVs... I usually truncate at ply 20. > >>It may be the replacement scheme. This particular position has a tendency to >>generate TT 'log-jams' where previous depth-preferred entries block out newer >>entries and therefore makes the search a little bit 'stubborn'... if you know >>what I mean. Try always-overwrite temporarily for a quick test. > >I think in general a replace always policy will be better, than a depth >preferred, even when it does not sound logical. Fabien mentioned, that Fruit can >store two scores/bounds for each position. I also experimented with this, but it >performed worse (in games especially, not so much in test positions) than the >normal scheme. I did not understand why, perhaps the implementation was buggy. I >tried hard to find any bugs, but did not see anything. > It sounds like it breaks the KISS principle... keep it simple/stupid >With my normal scheme (replace always, but also some space reserved for depth >preferred entries), Yace will solve Fine 70 in practically no time, when it uses >only 1000 hash entries. But I agree, that one should not tune hashing on Fine >70. Its interesting that you use mainly replace always. I have two tiers, so 50% entries are depth preferred and 50% are always-overwrite. With the Athlons I believe its cheap to read 64 bytes of memory so I might try using 4 slots (16 bytes each)... the first one being depth-preferred... the rest overwrite, then its 25%-75% which is maybe a better ratio. Thanks for your insights Dieter. Its always helpful. Ross
This page took 0.01 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.