Is it already a thing to consider the first part of the pv fixed when it is stable enough (let's say the first 25% of it when depth is at least 10 has not changed for the last 4 depth) and to start searching from the corresponding position (an apply that recursively).
The idea is to try to go deeper in a specific line.
Worth the try ?
Deeper search focusing on a specific line
Moderators: hgm, Rebel, chrisw
-
xr_a_y
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
-
Daniel Shawul
- Posts: 4185
- Joined: Tue Mar 14, 2006 11:34 am
- Location: Ethiopia
Re: Deeper search focusing on a specific line
With modern aggressively pruning searchers, it is probably already the case that very little time is spent if the PV move has no competitor.
If you do have two or more moves that are equally good, you probably want to search them equally anyway.
There is an idea forwarded for MCTS ("sufficiency threshold"), which suggests that if two or more moves score above a certain threshold (say 65%), randomly focus on one of them i.e. set exploration coefficient to 0 since you are winning anyway, and search one of them to deeper depths. If it turns out that that move is bad you can switch to the other move. You can try something like that in AB search too I guess.
If you do have two or more moves that are equally good, you probably want to search them equally anyway.
There is an idea forwarded for MCTS ("sufficiency threshold"), which suggests that if two or more moves score above a certain threshold (say 65%), randomly focus on one of them i.e. set exploration coefficient to 0 since you are winning anyway, and search one of them to deeper depths. If it turns out that that move is bad you can switch to the other move. You can try something like that in AB search too I guess.
-
Deberger
- Posts: 91
- Joined: Sat Nov 02, 2019 6:42 pm
- Full name: ɹǝƃɹǝqǝᗡ ǝɔnɹꓭ