Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash Collision

Author: Robert Hyatt

Date: 20:21:11 12/07/02

Go up one level in this thread


On December 07, 2002 at 11:42:54, David Rasmussen wrote:

>On December 07, 2002 at 11:25:44, Robert Hyatt wrote:
>
>>>
>>>I hadn't either until now. But I am pretty sure that Bob sais something about
>>>this happening once in 100 games or so.
>>
>>That was an estimate.  Think about the math.  You are collapsing some 2^168
>>positions into 2^64 hash space.  That is a _huge_ aliasing problem.  So the
>>probability of getting duplicates is not just a probability, it is an absolute
>>certainty...
>
>I know that. I am not saying that it will not happen. I know it will. But I am
>asking about the frequency. When doing math on these beasts people tend to just
>assume that everything is random. It isn't. The subspace visited by a 12-ply
>search with a 64 MB hashtable is not the same as the whole game tree. As I said:
>we don't care if there are signature collisions between two positions that are
>"far" from eachother. The probablility for it happening within the history of a
>seach with a hash table, is far less (apparently). And it is not simple to
>estimate how much. So I was asking about people's experiences of this frequency,
>because I didn't believe such naive reasoning anyway. If the theory isn't
>corresponding with experience, fix the theory.
>
>/David


The math tells the tale.  Searching 1M nodes per second, in a game that goes
for (say) four hours (forty moves in 2 hours) means 14400 seconds, which is
14 billion nodes.  You are mapping 2^168 positions into a 2^64 hash signature,
so it is not surprising that an error occurs here and there.

I don't remember the results any longer, but several of us ran some tests years
ago in a discussion on r.g.c, and found that 32 bits is useless, and that 64
bits has so few errors they can be ignored.  With recent testing I did that
suggests that a bad hash collision every 10K nodes has no effect on the final
score, we seem to be "ok".  :)




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.