![[tr]](/mwf/v12/flags/tr.png "Turkey")
Hi all,
think that there is a position and the engine iterate and search the position depth by depth,
in the depth 14, it spend 20 secs and it finds a move and gets alpha value 0.36 and this is PV. after finding PV it thinks it has good result for the position (0.36). for example it will spend for other moves 15 seconds more to go to depth 15 or remaing moves. if the engine gives litte greater value to alpha than real score (let say .36+0.5=.41) to get to depth 14 and prune the other moves faster (if there is no tactical move), do you think this will crush other things such trasposition table or the engine misses more important move for the position while giving extra point to PV ? and this is the right place to ask ? :)
thanks
think that there is a position and the engine iterate and search the position depth by depth,
in the depth 14, it spend 20 secs and it finds a move and gets alpha value 0.36 and this is PV. after finding PV it thinks it has good result for the position (0.36). for example it will spend for other moves 15 seconds more to go to depth 15 or remaing moves. if the engine gives litte greater value to alpha than real score (let say .36+0.5=.41) to get to depth 14 and prune the other moves faster (if there is no tactical move), do you think this will crush other things such trasposition table or the engine misses more important move for the position while giving extra point to PV ? and this is the right place to ask ? :)
thanks
![[au]](/mwf/v12/flags/au.png "Australia")
> if the engine gives litte greater value to alpha than real score (let say .36+0.5=.41) to get to depth 14 and prune the other moves faster (if there is no tactical move), do you think this will crush other things such trasposition table or the engine misses more important move for the position while giving extra point to PV ?
This sort of thing has been suggested from time to time. I don't see any "crash-causing" problems with it, but its value is debatable, and might be dependent on many other search-design issues. Also, there is the question of whether this should be done only at the root, or at all PV nodes (for instance, in your example, once the PV is found in the various depth 13, depth 12, ... depth 1 subtrees, should a similar delta be added to the score for pruning purposes?).
> and this is the right place to ask ? :-)
Maybe the Computer Chess subforum would be better. :)
I am trying to implement some sort of things sometimes like adding more point to real value, the engine prunes more and works a bit faster. I am not sure this type of pruning works well to find better move at all. and also this should be applied at the root of the tree, if not, then engine does not work properly. and also I think this needs many control at the root before calling subtrees. and also this extra point should not be used in the hash table (I think). I am not sure at all.
![[hu]](/mwf/v12/flags/hu.png "Hungary")
You're thinking along the right lines. This is a complicated topic and the answer will depend on a zillion other things in your engine. For example, if your engine already emphasizes the PV for other reasons, this is less likely to work.
Vas
Vas
yes I agree it's very complicated especially when and where to implement in the tree. I have tried some different things, it's really fast but I realized that it's not a good way to prune because of it needs so much control not to miss better move. and I saw that the engine may miss some important moves or this conflicts with other pruning tech. or eval function, and prune a move that it should not.
Regards...
Regards...
Right, these are just tradeoffs. Search (and also eval) are about tradeoffs ..
Vas
Vas
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill