Author: Robert Hyatt
Date: 15:06:57 09/28/00
Go up one level in this thread
On September 28, 2000 at 12:27:09, Carlos del Cacho wrote: >On September 28, 2000 at 00:18:58, Robert Hyatt wrote: > >>On September 27, 2000 at 12:10:00, Carlos del Cacho wrote: >> >>>On September 27, 2000 at 09:26:45, Robert Hyatt wrote: >>> >>>>On September 27, 2000 at 07:47:18, Bas Hamstra wrote: >>>> >>>>>Supposing no "lazy-errors" at all were made, does anyone know if there are >>>>>serious side-effects to lazy eval? >>>> >>>>None at all. Except that guaranteeing this is a bit hard. :) >>>> >>>> >>>> >>>>> >>>>>I one experiment that I yet have to repeat, it seems that NPS increases, but >>>>>Depth (as averaged over 300 wac positions) does not. >>>>> >>>>>I would like to know if others have seen alike or other problems with LE. >>>>> >>>>> >>>>>Regards, >>>>>Bas Hamstra. >>>> >>>>It has been used forever. As a classic trade-off between speed and accuracy. >>> >>>Just a question related with this. Since I implemented lazy eval in my program I >>>don't store the value returned by search in the hash table when there's a fail >>>high or a fail low. I just store beta or alpha instead. Is this correct ? >>> >>>Thanks in advance, >>>Carlos >> >> >>That is one way to do it. Storing values outside alpha/beta (referred to as >>fail-soft alpha/beta) also works... > >Yes, that was the approach I used before. But it seems to work worse now. >Suppose I get to a node where lazy eval gets applied. Backing up the tree I have >no idea it did so. Suppose I fall high. I can't store this score as a low bound >on the true score unless I store some info of the window used, because if >researched with other alpha-beta window score can get lower than this. >I'll keep thinking about it. >Carlos That is a cost of lazy eval. I don't really use the bound information anyway, so fail-soft is only useful for storing wider than normal bounds so that the hash entries are useful more frequently. But you obviously miss some of the 'widening' effect if you stop evaluating too soon.
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.