Author: Daniel Clausen
Date: 10:00:01 01/19/04
Go up one level in this thread
On January 19, 2004 at 12:28:39, Tord Romstad wrote: [snip] >I agree that the difference in speed probably isn't noticable (at least >not in a very slow engine like mine), but Like Sune, I think it feels >cleaner to view building a PV as a task separate from the search. I definitely don't try to convince other people here, but for the sake of completeness, here's what I think :) The score and PV are the _result_ of the search, return values so to speak. The best move is the very first move in the PV. Whether the search internally uses HTs, IID, PVS and whatnot, shouldn't change that. (apart from nasty effects, which can change the PV at times, but not the score) That has the nice side effect that I can get a PV even if I turn off HTs for debugging purposes. >I haven't seen the PVs Shredder produce, but in my own engine it seems >that the problem with wrong PVs is very rare. The first two moves always >seem to be correct, and I can only remember a handful of cases where there >was something obviously wrong later in the PV. I'm sure it happens a >lot more frequently than I notice it, though. Of course I don't study >all the thousands of PVs my engine prints. A difference between engines who retrieve the PV from the HT could be, that some verify whether the associated move in the HT is legal while others don't. (they all do during the search - or they should :) OTOH, when I tested with my old engine, it almost never happened, that the move from a successful HT-lookup was not legal, so I don't think that can be an explanation that some engines suffer from wrong PVs more than others. (_if_ there's really a difference) Sargon
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.