Author: J. Wesley Cleveland
Date: 17:35:41 11/12/02
Go up one level in this thread
On November 11, 2002 at 15:44:25, Robert Hyatt wrote: >On November 11, 2002 at 15:04:15, J. Wesley Cleveland wrote: > >>On November 10, 2002 at 21:29:43, Robert Hyatt wrote: >> >>>On November 10, 2002 at 21:15:07, Jim Bumgardner wrote: >>> >>>>Which of these strategies for "think on opponent's time" makes more sense? >>>> >>>>A) To only search the top-move from the principle variation. If >>>>the opponent makes that move, continue searching, otherwise reset and >>>>search again. >>> >>>This is the _only_ way to do it. I've explained this many times, but it >>>is probably time to go it again... >>> >>>Suppose you predict your opponent's move correctly only 50% of the time. >>>And it should be pointed out that this is a _low_ estimate from thousands >>>of observed games (via log files). This means that 1/2 of the time, you will >>>predict correctly and when your opponent moves, you have an instant response >>>ready. 1/2 of the time you get to think for free. >>> >>>Suppose you choose to search the top three moves instead of just the first one. >>>When your opponent has moved, you have spent 1/3 of the total time on each move. >>>You save 1/3 of the time. And that is worse than saving 1/2. If you only >>>search the top 2 moves, you will save 1/2 of the time, _if_ the move played is >>>one of those two, but occasionally it won't. >>> >>>It is really simple to see why searching only the best move is the right >>>idea. I could think of a few cases where I might vary this, such as where >>>my target time is 3 minutes and my opponent searches for 12 minutes. Do I >>>want to search one move for 12 minutes, or do I want to take a chance and >>>use 1/2 of that time (say) to search for an alternative best move? Tough to >>>say, and although I have tried such ideas many times, I have always come back >>>to searching what I consider the best move only. And since 50% is a low >>>prediction percentage, searching one move actually is even better than the >>>above pessimistic analysis. >> >>This stratagy is clearly the best until the total alloted time for this move is >>used, i.e. you would move instantly if the pondered move is made, as the effect >>of pondering is saving time for future moves, and the greatest expected saving >>is given by searching the expected moves. After this, at some point the >>advantage goes to searching other moves. > >Maybe. But remember, the effect is just a "deep think". IE I might spot >something >really good in 6 minutes, but not in 3. If I stop after 3 and go on to another >move, >I might not find the deep sac... > > >> A clear example of this is asymmetric >>time limits, e.g. you have 1 second per move and your opponent has 1 day per >>move. Here, you could easily blunder if you have not pondered the move the >>opponent actually makes. The point at which to switch depends on the expected >>value of searching the current move deeper compared to the expected value of >>possible time saved by pondering another move. This is not easy to calculate, >>and experiments would need to be done. > >I'd agree with that. I'm not sure what the time limit would be. But another >case is >when the opponent crashes. And you are pondering. I've pondered for nearly two >hours >in a couple of ACM events, and that might be better spent looking at more moves >of >course. But I am not sure where the 'break-out" point comes... One idea is on finishing a ply, if you have used the time alloted for the move, switch to searching all moves. Something I just (finally) realized is that it is not a choice between pondering the best move and the second best move, because you don't know what the second best move is.
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.