Author: José Carlos
Date: 09:00:57 11/24/02
Go up one level in this thread
On November 24, 2002 at 10:48:52, Robert Hyatt wrote: >On November 23, 2002 at 13:38:32, José Carlos wrote: > >>On November 23, 2002 at 12:53:36, Omid David Tabibi wrote: >> >>> >>>>Nodes to solution should be a better parameter. >>>> >>> >>>Provided that you have a pool of positions that can be solved within a >>>reasonable time! >> >> You should find them to provide useful data. Tree reduction doesn't mean >>anything if the solution is not found. Nodes to solution is definetly the >>important data. >> >> José C. > > >Maybe I am overlooking something, but as best I could tell, his new algorithm >solved more positions than R=2, at the same depth. _and_ the tree was smaller. >I don't see how that can be considered a bad way of reporting things. He ran >to fixed depth, which I personally think is very reasonable since it provides >something that can be repeated. He reported the sizes of the trees, which shows >that his approach searches a smaller tree than straight R=2. And he provided >the number of correct solutions showing that his approach is better than >straight R=2 also. > >To consider his data invalid, you would have to assume one of the following: > >1. His positions are _all_ zugzwang positions, so that any zugzwang detection >would make it perform better than straight null-move R=2. > >2. Somehow his search makes searching the PV much harder, and the non-PV moves >much easier, so that reporting the time to solution would somehow show larger >numbers for his new code, even though the time to complete a depth is lower. >(I assume the time to depth is lower since he used the same program for both >and I don't think his NPS will be affected by the algorithm significantly). > >I find either of those to be a bit hard to accept, and I concluded that what >he did worked for him. Whether it will work for the rest of us or not is >another thing... I agree. Maybe I didn't express correctly. My aswer was directed to the following statement: "Provided that you have a pool of positions that can be solved within a reasonable time!" What I meant is: if you use positions that won't be solved by either algorithm under the chosen testing conditions, the fact that the tree is smaller doesn't prove anything, as both algorithms will a bad move in that position. So it's better to find positions where a best move exists and measure time or nodes to find it. Of course, smaller tree plus more solutions is a positive result, but smaller tree alone isn't. José C.
This page took 0.11 seconds to execute
Last modified: Thu, 07 Jul 11 08:48:38 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.