Author: Stuart Cracraft
Date: 20:43:12 07/25/04
Go up one level in this thread
>This is not clear to me. >If you have a best move at the root the engine can play it. >It might be better than the best move of the preceding iteration. >(of course you have to make sure that you don't backup anything to the root >after the timeout, the score and the best move at the root shouldn't change >after that). ** This was one thing I didn't do. It does now. Thanks. >Only if you have no best move it should play the best move of the previous >iteration. Is this what you are doing? > Even though no search has ever not produced a best move without use of the previous iteration, I think I should put this in anyway. *** The real bug was found however to be in the time-recording code only surprisingly. *** This was saying it was taking longer than 5 seconds for a move when it was really taking almost exactly 5. I solved this with a pocketwatch and comparing what it reported with what the pocketwatch said. Imagine the surprise. Now the suite of 300 tested at 1 second per position is reporting as finishing in 267 seconds instead of 330, 340, 350, at 1 second, while the total solved is still the same. So that is great. The shorter time of course is due to the rapid solution of those positions with mates or draws taking less time. The other side effect of this fix is that nps has jumped up to 0.25M/s (node counted as entry with non-trivial exit in main search or quiescence search.) Thanks for your input!
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.