Author: Don Dailey
Date: 12:50:10 09/30/98
Go up one level in this thread
On September 30, 1998 at 15:33:33, Robert Hyatt wrote: >On September 30, 1998 at 12:10:44, Don Dailey wrote: > >> >>>>Which is pretty much how I do it. But my main point actually was how >>>>one should THINK about null move selectivity. Thinking of it this way >>>>is a great simplification and can make it easier to understand. >>>> >>>>My program actually does a null move search with a zero width window >>>>around beta, so I don't actually get to raise alpha. In lots of testing >>>>I did this was more efficient, I don't know if anyone else has tried >>>>this or finds it useful. I also do my test at the top of the search >>>>like you do but after making the move passed to the search function. >>>> >>> >>> >>>My null-search is also with beta-1,beta as the window, because I only care >>>about failing high, and wouldn't trust the null-move to "raise alpha" in >>>those few cases where I am not already doing a null-window search in the >>>PVS code. >> >>Hi Bob, >> >>I have the same concern. There were a few positions that caused problems >>and since the zero window is fast, there seems to be no justification >>for having to put up with the alpha problem. >> >>- Don > >I thought this was a moot point in your case anyway, since using mtd(f) >means *every* node is searched with X,X+1... ??? I use the score trick of using even numbers and having a single beta variable which is odd. There is not even an alpha in principle! But I have another program I'm working on too, not a 64 bit program and simplified. This is to be another cilk example in our next cilk distribution. I use pure alpha beta without even using aspiration tricks except for this null window test for selectivity. >almost all of mine are searched the same way. This "thinking" probably >goes back to the Slate/Atkin days when they talked of using the hash table >to "raise alpha" or "lower beta" if the bound wouldn't immediately cause a >cutoff. Way before negascout/PVS type algorithms... > >I Never did this funny-business with hashing, because it never seemed to be >100% safe... each time I tried it I would see fishy things here and there...
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.