Author: Georg v. Zimmermann
Date: 04:45:43 12/12/00
Go up one level in this thread
On December 11, 2000 at 15:33:33, Carlos del Cacho wrote: >On December 11, 2000 at 15:28:33, Georg v. Zimmermann wrote: > >>On December 11, 2000 at 15:00:36, Frank Phillips wrote: >> >>>What is the standard approach to selecting the (root) move in an iterated search >>>when time runs out part way through the next iteration at the next depth. Going >>>back to the pv for the last completed iteration sometime seems a waste, >>>particularly with hash tables when it could have taken less than a second to get >>>to that point and especially when the next incomplete iteration has a different >>>root move. Using the pv based on a partially completed search could be bad >>>since had it continued a refutation might have been found. Equally it could >>>have turned out to good……….. >>> >>>In my search I do not even start the next iteration is there is less than 50 >>>percent of the total allowed time left. >>> >>>Frank >> >>This is an interesting problem and I have run a number of experiments, >>unfortunately with unclear result. In principle the search time to finish a >>given interation should be "more valuable" since the closer bounds will result >>in more cutoffs resulting in a higher n/s. Or is that reasoning flawed ? >> >>Georg > >I don't think that's a good idea. You will be searching last iteration's PV >first. If at the time you stop your thinking the PV move is differemt you know >that that move is better at that ply that the ones previously considered. So you >should play that move IMO. > >Carlos Argh sorry, my formulation was unclear. Obviously you should always use whatever search info you got, the only exception being that its not a good idea to use a value when stoped in the qsearch I guess. I was talking about how to distribute the search time. And if it is right that the time spend to finish an iteration is "worth" more than the one used to start it, then it would make sense to try to search all first moves as often as possible. Hope that makes sense :) Georg
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.