Author: rasjid chan
Date: 01:23:20 06/22/05
Go up one level in this thread
On June 21, 2005 at 15:27:10, rasjid chan wrote:
>Dieter,
>
>Some of the things here by Vasic and you are actually advanced stuff for me
>and I am not yet at the level to completely handle them yet. Just like Vasic's
>mention of being optimal.., I need time to consider those subtleties, but I keep
>things in mind that there are such things to be alert to when it's time.
>
>I can sense both of you did more formal reading but I get most tips here after
>I come about 2 years back.
>
>Just as an example of the big gap in my knowledge.I was fixing my rep3 code
>after being reminded by Vasic about the correct 0 score and I got stuck; it
>seems not that simple! So after thinking, I found it could be the graph tree
>interaction stuff of Breuker you refer to. It may be wrong! :-
>
> This is for the sake of rep3 theoretical accuracy.
>
> updatehash();//for the node B after makemove() below
> if (rep3()){
> best = 0;
> .....
> // let this node B repeats A down below.
> }
>
> makemove();
>
> Say rep3 is triggered. Then all nodes searched after B will inherit
> dependency until we leave A. Had B being reached through another path w/o A
> below we don't have the best = 0. So basically subsequent sub-branch searched
> will be corrupt until we leave node A. I may be wrong as I just thought of it
> an hour ago.
>
> I will make copy of my ab_search_w/o_storehash() to disable all hashing(not
> probehash) until search leaves node A. I think just return ply of A from
> rep3().
>
> Don't laugh at what I am going to do. I am only experimenting, not weighing
> efficiency yet.
>
> Vasic,
> Incidentally, just to let you know. BU fail-soft could pass inconsistencies
> from QS to horizon depth/pre-horizon depth when we have QS generate check. I
> remember I need always to gen-check for 1st ply of QS and maybe also 2nd ply.
> The move set of QS changes. Also a need to keep 1 bit _qs_gen_check in HT.
> You'll know what I mean.
>
> Best Regards
> Rasjid
My mistake again, hope this is final.
I am just trying to first get the stict conditions correct.
Node B has a rep3 and whatever the final score (need not be 0) be for B has
dependency. This dependency propagates to the parent node only and whatever the
final score of this parent be inherits the dependency. This propagates all the
way down to root. So strictly for HT purity all nodes along root_node -> A -> B
should avoid hashing; ie. even the root search for this iteration should not be
hashed.
But this inheriting of dependency at root may not happen at every iteration as
node A may be pruned. Even changing the move ordering may avoid dependency.
Best Regards
Rasjid
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.