Author: Robert Hyatt
Date: 16:45:56 11/22/01
Go up one level in this thread
On November 22, 2001 at 16:00:43, Mike S. wrote:
>On November 22, 2001 at 09:42:36, Robert Hyatt wrote:
>
>>On November 22, 2001 at 05:27:05, Gordon Rattray wrote:
>>
>>>The Fritz GUI analyses games ("Full Analysis") by starting at the end of the
>>>game and retracting moves. How does this compare to going forwards? Does it
>>>produce better results? (...)
>
>>Here is the idea...
>>
>>If you start at the end of the game, you load the hash table with stuff
>>that will help as you search at earlier moves... with the "idea" that
>>earlier analysis will be more accurate since it will have access to these
>>scores.
>>
>>It doesn't work however.
>>
>>IE pick three points in the game, (a) where a key mistake is made, (b) a
>>position further into the game, and (c) a position near the end where the
>>program can see that it is lost. As you search backward, when you reach
>>(b) the search might well _still_ see that it is lost, because of the persistent
>>hash entries that help. But when you back up past (b) eventually the
>>hash entries get replaced, and you "lose the key scores". You don't find the
>>_real_ place where you screwed up (a), instead the score seems to drop at
>>(b) which is the wrong place.
>>
>>Since neither way finds the actual mistake, I don't like the back-to-front
>>approach because if you do search front to back you will find the "mistake"
>>at a different place, which is nothing more than confusing.
>
>I know what you mean, but I have seen that engines behave very differently at
>this. Maybe it's because they do the replacement of hash entries in different
>ways (I suspect some remove entries although there's still a lot of free space
>in the tables, because I usually stay only for some seconds on a position), or
>maybe some are especially equipped for that analysis method.
Doesn't matter, when you think about the math. Either you keep all the
positions from the first searches, which means newer searches (going backward)
will have no table space to write into, or vice-versa...
My feeling is this: Where a critical mistake is made, either the engine will
directly see it or it won't. If it won't, then it is most likely that going
backward will let the engine spot the problem earlier, but _never_ early enough
to find the _real_ mistake. And if it can't find the real mistake, I'd rather
not have it producing analysis that is based on random table overwrites as to
what it sees, and when...
>
>I recommend the Hiarcs 7.32 engine for that purpose. You should see how it acts
>in a ChessBase GUI while retracting in analysis mode. I have often seen it
>actually "carry" the exact scores upwards through the notation until it found a
>better alternative. I don't recall seeing it re-issue a mistake so to speak, if
>I rectracted from a position where the effect of it was visible.
>
>I wonder if this has to do with the low node rate of Hiarcs, compared to other
>engines (except The King). Maybe it's easier for it to keep the scores because
>of this.
possibly, but what happens on 30 minute searches? Can't store those no matter
how slow it is...
>
>Regards,
>Mike Scheidl
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.