Hello!
I just downloaded the latest free version of Komodo 12 to try to disable the Null Move Pruning option, so in my mind, I supposed that for depth 2, it will need 20*20= 400 nodes However I arrived to depth 3 with nodes 78 and depth 4 with 193.
So I'm wondering what is the role of the Null Move Pruning option, I thought that it will search every single position even if the moves were disastrous.
Thanks for your help!
Null Move Pruning
Moderators: hgm, Dann Corbit, Harvey Williamson
-
ThomasMiller
- Posts: 25
- Joined: Mon Nov 02, 2020 5:14 pm
- Full name: Thomas J. Miller
Re: Null Move Pruning
You have disabled just Null Move Pruning not all the other algorithms Komodo uses. To see 20 nodes for depth 1, 400 nodes for depth 2, etc you would need Komodo to use Minimax, no quiescence search, no transpositions tables and so on.wihdv wrote: ↑Wed Feb 24, 2021 3:43 pm Hello!
I just downloaded the latest free version of Komodo 12 to try to disable the Null Move Pruning option, so in my mind, I supposed that for depth 2, it will need 20*20= 400 nodes However I arrived to depth 3 with nodes 78 and depth 4 with 193.
So I'm wondering what is the role of the Null Move Pruning option, I thought that it will search every single position even if the moves were disastrous.
Thanks for your help!
-
hgm
- Posts: 27701
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Null Move Pruning
The most important form of pruning in engines is alpha-beta pruning. And they usually do not provide an option to turn that off.
With perfect move ordering, a 2-ply alpha-beta search would only have 1*20 + 19*1 = 39 leaf nodes: only the PV move would get all 20 replies searched, but the other 19 moves would already have been refuted after the first reply, and never search any others.
This is an under-estimate, though, because engines will also do a Quiescence Search after they have played the indicated number of plies, and some of the replies to the PV would expose a Pawn to capture by a Bishop or Queen, and the QS nodes are counted too. Furthermore, internal nodes of the tree would probably also be counted, so you have to add 21 more for the root and the nodes after the first ply. The 39 are only the leaves of the full-width search.
With perfect move ordering, a 2-ply alpha-beta search would only have 1*20 + 19*1 = 39 leaf nodes: only the PV move would get all 20 replies searched, but the other 19 moves would already have been refuted after the first reply, and never search any others.
This is an under-estimate, though, because engines will also do a Quiescence Search after they have played the indicated number of plies, and some of the replies to the PV would expose a Pawn to capture by a Bishop or Queen, and the QS nodes are counted too. Furthermore, internal nodes of the tree would probably also be counted, so you have to add 21 more for the root and the nodes after the first ply. The 39 are only the leaves of the full-width search.
-
RogerC
- Posts: 41
- Joined: Tue Oct 29, 2019 8:33 pm
- Location: French Polynesia
- Full name: Roger C.
Re: Null Move Pruning
Hi, null move pruning is not the only reductions that recent engines use to avoid losing time on useless evaluations of variations.wihdv wrote: ↑Wed Feb 24, 2021 3:43 pm Hello!
I just downloaded the latest free version of Komodo 12 to try to disable the Null Move Pruning option, so in my mind, I supposed that for depth 2, it will need 20*20= 400 nodes However I arrived to depth 3 with nodes 78 and depth 4 with 193.
So I'm wondering what is the role of the Null Move Pruning option, I thought that it will search every single position even if the moves were disastrous.
Thanks for your help!
Engines use also reductions like Late Move Reductions (LMR), Fail-High Reductions (FHR), RankCut, Razoring, Futility Pruning, Quiescence Search, Multi-Cut and ProbCut. This is why those reductions push evaluations going very fast in deep search (30 plies in a few seconds) and make recent years engines very strong.
Best regards.