Author: Robert Hyatt
Date: 14:36:39 08/10/01
Go up one level in this thread
On August 10, 2001 at 16:04:18, Artem Pyatakov wrote: >I just implemented pondering in my program, but it seems that this has a >negative effect on average search depth in games. Here is my reasoning: > >1) My engine is not very strong, so especially in games against humans, the >percentage of the times when it ponders correctly in "indecisive positions" is >low. >(Aside question: What approximate percentages do programs like crafty have?) Typically 50% at least. Much higher against very strong humans. > >2) So, let's say the ponder was incorrect. Wouldn't the ponder process have >ended up loading a lot of "junk" into the hash table and the history table? > Yes. But those positions are very close to the actual positions you will search if the person plays something else. Think about "transpositions". You expect me to play Nc3 and you think h6 is best followed by my playing Bg2. But I actually play Bg2. When you search h6 Nc3 you will hit that result and not need to search it. Works very well... >Does this even sound reasonable? Has anyone run into this? Or is this most >probably a bug? Any suggestions/comments are appreciated. You _can_ tell that the entries from the ponder search are not from the real search, by your "aging" mechanism, right? IE if you ponder for 10 minutes and load up the table, but the opponent plays an unexpected move, you should "pretend" that all the entries in the table are empty as far as overwriting them goes, although if you get hits on them at lookup time, they should be used... If you get this wrong, then yes you can hurt hashing seriously. > >Thank you. > >Artem
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.