I did not try it, but I just got a funny idea.
What if different threads in the search had different null move pruning values?
Let's suppose the initially calculated reduction is 10 plies.
Perhaps create a bell curve where one thread is zero plies up to one thread with 21 plies with most threads using about 10 plies using a bell curve. If you have 64 threads or more, it might prove useful, I think.
It seems like it might spot missed dangers and might also see deep tactics. Of course, it would be hard to debug and very hard to repeat the searches.
It just seemed like a fun thing to try.
It would need to use a timed search, since the threads with very little reduction would take a long time to finish a ply
Odd idea - null move spectrum "on a curve"
Moderators: hgm, chrisw, Rebel
-
- Posts: 12684
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Odd idea - null move spectrum "on a curve"
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
-
- Posts: 223
- Joined: Tue Apr 09, 2024 6:24 am
- Full name: Michael Chaly
Re: Odd idea - null move spectrum "on a curve"
This was tried in stockfish with no elo gained from this - as well as for other search heuristics (RFP, ProbCut, etc).
The only thing which really works is increasing reductions with thread count - but it does so uniformly for all threads, so isn't really the same.
The only thing which really works is increasing reductions with thread count - but it does so uniformly for all threads, so isn't really the same.