Author: Bo Persson
Date: 00:43:50 04/01/00
Go up one level in this thread
On March 30, 2000 at 18:34:17, Robert Hyatt wrote: >On March 30, 2000 at 15:47:03, Inmann Werner wrote: > >>On March 30, 2000 at 15:39:09, Peter Fendrich wrote: >> >>>On March 30, 2000 at 15:04:09, Inmann Werner wrote: >>> >>>>On March 30, 2000 at 11:07:16, Robert Hyatt wrote: >>>> >>>> >>>>>Here is mine: >>>>> >>>>>1. hash table move. >>>>>2. captures that don't appear to lose material using a SEE procdedure, >>>>>ordered from biggest gain to equal exchanges. >>>>>3. 2 killer moves. >>>>>4. up to 4 history ordered moves (history heuristic) >>>>>5. rest of the moves. >>>> >>>>question to 5) >>>>here is the rest of the non capturing moves and the "loosing capture" moves. >>>>Which of them should be searched first? >>>> >>>>IMHO the non capturing moves. >>>> >>>>Werner >>> >>>I don't think you should order them at all... >>>When the program reaches this point it will probably not find a fail high for >>>the current node and the sorting will only cost performance without giving much >>>in return. >>>//Peter >> >>Excuse, but I do not agree. >>Why should a good positional move not produce a fail high? >>And i do not sort. I only give the moves "values" at generation time. In search, >>I only look at the first 9 moves in an ordered way, the rest i pick at random. >>My question is: When I produce the "loosing captures" (together with all >>captures), I can give them a small positive or a negative value (without cost). >> >>I give them a negative value, and it works a little better than otherwise. >> >>Werner > > >The reason is this: If you spend much time sorting, and (as in Crafty) 92% >of the fail highs happen on the first move searched, by the time you get thru >the captures, the killers, and the history moves, the probability of a fail >high is _very_ remote. How much time are you willing to invest to get that >occasional quick fail high? Hard to say what the right answer is, there... Exactly! Look at it this way Werner: The odds that there even *is* a good move is slim to none. Tried in any specific order, you will probably have to test them all anyway. So put the effort in working thru the list and not in trying to separate bad from slightly worse. Bo Persson bop@malmo.mail.telia.com
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.