Author: Uri Blass
Date: 00:47:16 11/12/02
Go up one level in this thread
On November 12, 2002 at 02:48:19, Sune Fischer wrote: >On November 11, 2002 at 17:39:50, Uri Blass wrote: >>>If we assume x is 50 like Bob's example, the first part of you equation would be >>>A% times 50% times 90% CPU, looks like that will be hard pressed to compete >>>against Bob's 50% times 100% CPU. Of course the assumption is that A=100, so you >>>only have 10% CPU to burn on the remaining 50% of the moves. You need 50% >>>efficiency out of those 10% to compete with Bob, and you have only moves that >>>failed low to search, I'd almost say that is impossible. >>> >>>-S. >> >>1)I assumed that x is 60% in my example. > >So we agree, there is one move with 60% probability of being >the best, and all the remaining moves cannot sum to more than 40%, for >a total of 100% of course. > >>2)I do not understand what you say here. > >I don't blame you, it wasn't very elegantly put. >Hopefully this time I've made myself understandable :) > >>I will explain again what you can practically >>do in your search: >> >>You can start with 90% time for the expected move >>and 10% for the rest of the moves with possibility to >>change it based on the evaluation(if you find >>during the search that the move that you expect >>is a bad move you can give more time >>for other moves and you can also give more time >>for other moves if you find that the score >>of other moves is the same as the score of the move that >>you expect. > >Yes you can, but you then get 90 efficiency on the 60% expected move, >that's 0.6*0.9+(something), this (something) must be something less than 40 >times the 10% that remains, right? >So 0.6*0.9+0.4*0.1=0.54+0.04=0.58 which it less than Bob's 0.6*1.0=0.6 >Further more it is an upper limit that assumes you are always able to pick the >correct second move, most probably you won't be 100% correct here, let's say you >are only right half the time (or equivalently: that you distribute the remaining >10% CPU power over more than 1 move), that comes to: >0.6*0.9+0.4*0.1*0.5=0.56 which is even less, but also more realistic. > >The only move that can give you 60% return on invested CPU power is the PV move, >therefore pondering that is the most effective. > >What to do when we _don't_ have a pv-move is a completely different question >(type B moves). > >I guess it is also possible that one might "detect" when the PV-move is among >those 40% misses, and thus effectively increase CPU returns, I think this is >really what you are talking about. Yes I may suspect part of the cases when the moves is in the 40% mainly when the main move fail low but also when I see another move fail high again and again even if the score suggest that the move that fail high again and again is 0.01 pawns lower than the best move. However I don't think I would waste any power >(even 10%) on searching other moves until there is some indication the PV-move >is a bad choice. The point is that you also may suspect that the pv move is a bad choice if you see another move fail high. Bad is relative to other moves. 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.