Author: Robert Hyatt
Date: 06:43:21 09/14/99
Go up one level in this thread
On September 14, 1999 at 05:56:30, Bas Hamstra wrote: >Hi Frank, > >I meant "internal iterative deepening". So when the original rootalpha and >rootbeta are still intact and there is no best move from the hashtable, you do a >search to a shallow depth just to get a good move. > >My question is: what does it buy you exactly? So to see it won't cost you much >(it only will do a couple of extra shallow searches), but does it do any good? > > >Regards, >Bas Hamstra. > I also answered the wrong question. Internal iterative deepening doesn't cost me much on any test... and generally saves me about 10%. I usually use the Kopec test as it has half tactical, half non-tactical. And overall IID gives me a 10% speedup on that test. It is _particularly_ good when you search, get a fail-high at the root, but can't get a real score because you fail low. At the start of the next search you might have no hash moves at all and that will totally kill the search. IID fixes this nicely. Note that on a given position you ought not do more than a dozen IID searches at most... it just doesn't happen very often unless you are doing it in places where you shouldn't.. > > >On September 14, 1999 at 05:10:39, Frank Schneider wrote: > >>Hi Bas, >> >>On September 14, 1999 at 04:07:18, Bas Hamstra wrote: >> >>>It seems I got PVS working. In comparison to simple alphabeta I get mixed >>>results. I have seen positions where PVS does 20% less nodes than ab (starting >>>position). However I have also seen (tactical) positions where PVS does do more >>>nodes than ab. I am trying to find out how that's possible, but didn't have >>>enough time. >>> >>>The impression is that pvs is just more sensitive to good moveordering than ab. >>> >>>I see that iterative deepening in Crafty. How important is it? >>Iterative deepening is used in almost every program. The main effect is, >>that by searching a position to depth n you get a very good moveordering >>for the subsequent search to depth n+1. >> >>There is also an additional technique called 'internal iterative deepening'. >>It is a little enhancement which can save time in some (rare) positions. >> >>Using iterative deepening is a must, internal iterative deepening is >>nice to have. >> >>Frank >> >> >> >>> >>> >>>Regards, >>>Bas Hamstra.
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.