Author: Robert Hyatt
Date: 21:26:00 09/10/99
Go up one level in this thread
On September 10, 1999 at 18:11:01, Eugene Nalimov wrote:
>Maybe the difference between Bob's and Ed's experimental results can be
>explained by the fact that different flavours of Alpha-Beta is being used - PVS
>in one case, aspiration search in other?
>
>Eugene
that is possible. PVS is a good bit faster than plain aspiration search,
but then the bounds are always null-width. Which increases the likelihood
(I guess) that the scores < alpha are meaningless. But in any case,
mathematically, the scores < alpha can not be sorted in any meaningful
way. If it is faster, it is faster for some other reason, because such moves
are _clearly_ more or less randomly chosen... Perhaps they get overwritten
often enough in the hash table that they don't crush the search?
>
>On September 10, 1999 at 17:46:45, Ed Schröder wrote:
>
>>On September 10, 1999 at 16:31:27, Robert Hyatt wrote:
>>
>>>On September 10, 1999 at 15:51:44, Ed Schröder wrote:
>>>
>>>>
>>>>I do the bestmove calculation BEFORE I do the A/B check.
>>>>
>>>>Thus...
>>>>
>>>>1) For each new ply in the search: bound=-INF; bestmove=0;
>>>>2) For each move on that ply: if (score > bound) { bound=score; bestmove=move; }
>>>>3) try A/B
>>>>
>>>>>What do I not understand about your idea?
>>>>
>>>>I don't know, hope it is clear now.
>>>>
>>>>>IE you aren't using PVS/negascout/aspiration search???
>>>>
>>>>Aspiration search.
>>>>
>>>>Ed
>>>
>>>Please check the article with "here is real data". I tried the above on several
>>>positions and it behaved as I expected... Have you confirmed that this is
>>>actually faster (or slower) for you than not trying to find a best move on
>>>fail-low positions? IE for me it was quite a bit slower, sometimes 20%...
>>
>>I have tried all possibilties in the past. I am happy with my system.
>>
>>Ed
This page took 0.01 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.