Author: Robert Hyatt
Date: 17:23:53 08/22/01
Go up one level in this thread
On August 22, 2001 at 15:53:38, Miguel A. Ballicora wrote: > >You do this even if you are running out of time and with a big advantage >already? Yes. In most cases, each root move takes a fraction of a second, after the first has been searched to get a score. Which means for most cases, this doesn't influence time usage at all. > That is what I am worried about if I decide to complete the analysis >of a root move. It if is good that it is worth anaylzing, it might take some >time, always less than the "main" move though. If it is fast, probably it was >not worth analyzing!. But I can see the point: every time a root move has been >interrupted there are wasted nodes. That is a fact. We better complete those >moves AND allow a bit less time to think per move to compensate. >But I would not do that if I am short of time and a piece up. What do you think? > If you want to make an exception, that is probably a good one. I just chose to "keep it simple, stupid" and do it every time... :) >Would not be a safer idea to have to different time deadlines? >t1 --> first time (the regular): time is up if we are analyzing the main move. >t2 --> (t2 > t1) if I am analyzing any other root move, I do not stop when the >time is > t1, I just continue, but t2 would be an "emergency" time out, in which >I _have_ to stop no matter what, in any situation, for instance I consumed too >much time because the search exploded and I am too close to lose >on time. Generally, this should never happen but it will guarantee some >"accidents" that makes me burn a lot of time in a position that is not worth it. I have that. I can use up to 2x the normal target trying to do what I explained, ie I won't burn forever. If the target is 3 minutes, I am going to move after 6 no matter what unless I get a fail low... >(i.e. I am winning anyway). >t2 could be adapted according to the evaluation after the main move was >analyzed. > >I am doing something similar if I interrupt in the PV move and the eval that it >returned is a fail low. That is bad news even though the search was not finished >so I continue the iteration, but until I find a move that does not fail low or I >finish the iteration or I reach the "emergency" deadline. > >This is not completely tuned up, but I think that I could use it along your >aproach. Whay do you think? > >Regards, >Miguel
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.