Author: Herman Hesse
Date: 14:20:24 07/30/99
Go up one level in this thread
On July 30, 1999 at 08:56:56, Robert Hyatt wrote: >On July 30, 1999 at 05:09:08, Herman Hesse wrote: > >>On July 30, 1999 at 02:36:56, Bruce Moreland wrote: >> >>> >>>On July 30, 1999 at 00:28:06, Ed Schröder wrote: >>> >>>>Search the first iteration without alpha/beta. >>>> >>>>Ed Schroder >>> >>>This would hurt your "game in one second" performance :-) >>> >>>bruce >> >>He told you all once. When Rebel was released able to give senseful 4-move lines >>for the first four moves. Four moves sorted into order. Only way was first >>iteration(s) without alpha-beta. No person ever commented. >> >>Now he told you twice. Only comment is a weak joke. >> >>Your forum stultifies when on-topic reigns. The off-topicals were the ones that >>saw and said so. Only dullness remains. Mr Schroder is sad about this. Possibly >>why he drops ideas in, hoping for souls to respond. >> >>Low iterations without alpha-beta lead to more senseful move ordering from hash >>when the PV is going to change in later iterations. This can save time. All >>spelt out so even the slow ones can see. >> >>Herman > > >Sorry, but this is _dead_ wrong. Otherwise we would just do the 'low iterations >without alpha/beta' and then play the best move. Check out Rebel on any test >suite. I'll bet you find it changing to the _right_ move at extreme depths and >not at depth 1 or 2 or 3. > >Doing a 1/2 ply search without alpha/beta can help a bit with ordering the root >moves, but _only_ a bit. Because all you learn is which moves seem to be best >for 1-2 ply searches. (no, you won't do 3-4-5 ply searches at the root without >alpha/beta and finish them in reasonable time). However, what is best at 2 >plies has _nothing_ to do with what is best at 10, 11 or 12 plies... > >This is an _old_ idea. Ken Thompson did a 2 ply search without alpha/beta in >Belle to order the ply-1 move list. I used his idea in Cray Blitz (I had the >ability to use a 'true-value' search at the root by resetting alpha/beta to >+/- infinity for each root move to get a score for each, useful for debugging >and other things). It isn't 'new'. It isn't particularly 'great'. And it >absolutely does _not_ let you see that all moves but one lead to mate. Because >they all (but one) might lead to a mate in 10, and you won't see that with a >1-2-3 ply search, generally. I amde a good, on-topic chess post here. It was with ideas. Mr Moreland said he deleted it. This was a nonsense deletion. Mr Moreland should justify or restore. I am quite sure Bob Hyatt can take care of himself. Herman
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.