Author: Dan Honeycutt
Date: 14:16:37 05/06/04
Go up one level in this thread
On May 06, 2004 at 15:47:05, Robert Hyatt wrote: >On May 06, 2004 at 15:32:11, Dan Honeycutt wrote: > >>On May 06, 2004 at 15:05:15, Robert Hyatt wrote: >> >>>On May 06, 2004 at 14:31:08, Gerd Isenberg wrote: >>> >>>>On May 06, 2004 at 11:58:11, Robert Hyatt wrote: >>>> >>>>>On May 06, 2004 at 11:38:44, Dan Honeycutt wrote: >>>>> >>>>>>Yet again I apologize for asking a question which I'm sure has been asked many >>>>>>times before, but could someone explain the difference between a fail-hard and a >>>>>>fail-soft and how does is affect a PVS search? >>>>>> >>>>>>I made the guess that, ie, if (score >= beta) if I return score it's a fail-hard >>>>>>and if I return beta it's a fail-soft. It would seem that what I return doesn't >>>>>>so much matter as what I put in the hash table. If I put in the (possibly) >>>>>>higher value of score then I have a higher lower bound and a greater chance for >>>>>>a hash cut if this position arises again. >>>>>> >>>>>>Thanks in advance. >>>>>>Dan H. >>>>> >>>>>That is backward. But you have the right idea. Fail-hard never returns a value >>>>>outside the initial alpha/beta window. Fail-soft does. >>>> >>>>I often confuse this hard/soft definitions too - i have the wrong mnemonic >>>>trick. Intuitively i found it harder if i jump outside a window ;-) >>> >>> >>>It is just as intuitive as where you store a lower bound but flag the position >>>as an UPPER bound position. Makes sense after a lot of thought, but it still >>>leads to confusion... :) >> >>Are the there known implications on the search from doing using fail-soft vs >>fail-hard? From a tiny amount of testing it looks like my engine does best >>using a fail-soft (outside the window) lower-bound (returning beta) and a >>fail-hard (inside the window) upper-bound (returning alpha). I'm sure I have >>something backward in that last sentence - see my reply to Anthony below for >>what I'm trying to determine. >> >>Thanks. >>Dan H. > >No problems I am aware of. With PVS it doesn't matter much since almost all >searches are done with a null-window... Would seem what you hash is more important than what you return. ie with a window of 1, 2 I get back a score of 50. The move at the previous ply fails low whether I return 2 or 50. But later I come to the same position with a window of 25, 26. If I have hashed the 50 I'll get a cutoff but if I hashed the 2 I won't. Dan H.
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.