Author: Robert Hyatt
Date: 18:57:07 04/25/01
Go up one level in this thread
On April 25, 2001 at 14:21:12, Bas Hamstra wrote: >On April 24, 2001 at 20:44:45, Robert Hyatt wrote: > >>On April 24, 2001 at 15:19:42, Bas Hamstra wrote: >> >>>On April 23, 2001 at 23:19:39, Robert Hyatt wrote: >>> >>>>On April 23, 2001 at 11:33:52, Bas Hamstra wrote: >>>> >>>>>On April 23, 2001 at 11:03:32, Ulrich Tuerke wrote: >>>>> >>>>>It is easy to check out where it goes wrong, by comparing the values with and >>>>>without FP. Values are in my case mostly different when the capture is checking. >>>>>Also in the endgame it is risky, when for instance a pawncapture brings the king >>>>>out of the "square of the pawn". >>>>> >>>>>In my case it does not matter much, because you will get a cut anyway the next >>>>>ply. It saves a call to ab/evaluate, that's about it. I can't see how that gives >>>>>enourmous gains. And it doesn't in my case. >>>>> >>>>> >>>>>Bas. >>>> >>>>What it does is cut the total q-search nodes by 1/2 at least. That is a pretty >>>>significant savings... >>> >>>Not in my case, no. Skipping SEE losing captures cuts 50%. Then, if I want >>>reasonably accurate scores, futility cuts another 15% or so. However I do not >>>consider any checking capture as futile. >> >> >>I don't follow. Skipping SEE losing captures cuts 50% is what you said? That >>is what I said.. the q-search gets 50% _smaller_.. >> >>?? >> >>Bob > >No, wait. YOU said futility pruning cuts 50% qnodes. *I* said it gives only 15% >or so but pruning SEE losers away cuts 50%. Look above :-) > >Bas. Note that the definition of "futility" I used was "losing captures that can't bring the score back into the window". Not the "futility pruning" applied near the frontier nodes. Se my use of "futility" was simply captures that are futile to search since they can't possibly produce a score inside the A/B window.
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.