Author: Robert Hyatt
Date: 16:56:19 10/29/03
Go up one level in this thread
On October 29, 2003 at 15:22:52, Dan Andersson wrote: > It would be interesting to know the limit where it becomes a real bother. And >what other factors might kill the search. If one corrupts the information stored >in certain ways like depth and result, what happens? > And if one had a more random approach like a P chance for a wrong match would >this less deterministic condition disturb the search more? > >MvH Dan Andersson The question I tried to answer was "how many collisions are too many?" It turns out the search is _very_ forgiving. Particularly if you use PVS. Most of the table entries don't have wild scores at all, just >beta or <alpha type values. As a result, false hash matches don't crush things anywhere nearly as bad as one would expect. I'll publish the results of this test before long, as its eye-opening, and it pretty well suggests that all the worrying we do about 32 vs 64 bit hash signatures is a wasted exercise. 32 bits will not give you a hash collision every 10 nodes, for example. I'm not sure even 16 bits is that bad, but this is another question my current testing will answer. (how many bits is enough to keep the collision rate above the critical threshold?) That's an interesting question.
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.