Author: Uri Blass
Date: 16:02:44 09/09/01
Go up one level in this thread
On September 09, 2001 at 18:43:48, Dave Gomboc wrote: >On September 09, 2001 at 18:11:51, Uri Blass wrote: > >>On September 09, 2001 at 17:28:38, Gian-Carlo Pascutto wrote: >> >>>On September 09, 2001 at 17:17:50, Uri Blass wrote: >>> >>>>Singular extensions can save a lot of plies only in positions when there are a >>>>lot of singular lines when there is a big difference between the best move and >>>>the next best move. >>>> >>>>In the releavant position it simply does not happen and if you search through >>>>the logical lines you find that most of the moves of both sides >>>>are not singular. >>> >>>Then explain why my SE crafty has a 3 ply advantage here. >> >>I said most of the moves and not all the moves so 3 plies difference make sense >>but not 20 plies difference. >> >>> >>>Two things: >>> >>>a) computers do not search logical lines >> >>The illogical lines are simply nodes that we can ignore and get the same >>conclusion. >> >>>b) there is no need for a big difference. my implementation will >>> already extend on 0.25 pawns, or 0.50 pawns depending on the >>> situation. >> >>In most of the cases the difference is even smaller than 0.25 and if you extend >>for a difference of 0.25 then a move that is 0.25 worse is not going to get >>extended from the root and you are not going to see that it is losing 2 pawns. > >Remember that while we call them "singular extensions", they actually did them >in cases which were not strictly singular. For instance, when there were only >three good moves, they extended all of them, though not to the same degree as >when there was only a single good move. > >Dave I thought that it is correct only about Deep blue. Do you say that it was also Deep thought? I also doubt if all these extensions were productive and it is possible that they believed that it should be productive at long time control and were simply wrong. I suspect that they had no time to test if these extensions really help relative to not doing it because they were too busy in implementing the new extensions in hardware. Uri
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.