Author: Robert Hyatt
Date: 10:08:18 02/26/02
Go up one level in this thread
On February 26, 2002 at 12:26:11, Vincent Diepeveen wrote:
>
>first of all it only got 11 to 12 ply. Secondly it is not deeper
>at all. it is the same depth. It only means sometimes you *might*
>prune away something which DF sees at 12 which DB sees at 11.
>
>In principle minimax as well as nullmove as well as alfabeta
>as well as hashtables give
>for a depth n the same score back to the root.
I'm not going to go into the "depth" issue again.
however, your second statement is _wrong_. alpha/beta provably gives the
same score as minimax. Null-move does _not_ and it is trivial to prove that
it does not, because it reduces the depth on selected branches while alpha/
beta does _not_ do that.
Null-move introduces errors into the tree, period. The gains seem to offset
the errors or we wouldn't be using it. But to say that it gives the same score
back is simply nonsense. I have done _many_ searches with Crafty with and
without null-move and I _know_ how the score changes...
Here is a 7 ply search with null-move:
7 0.70 0.20 1. e4 e6 2. Nf3 d5 3. e5 Nc6 4. d4
7 0.87 0.24 1. e3 d5 2. Nf3 Nf6 3. Nc3 Bf5 4. d3
7 1.01 0.29 1. Nc3 d5 2. d4 Nf6 3. Nf3 Bf5 4. Bf4
(3) 7-> 1.08 0.29 1. Nc3 d5 2. d4 Nf6 3. Nf3 Bf5 4. Bf4
and here is the same 7 ply search without:
(2) 7 2.81 0.20 1. e4 e6 2. Nf3 d5 3. e5 Nc6 4. d4
<HT>
7 4.00 0.29 1. d4 Nf6 2. Nc3 d5 3. Nf3 Bf5 4. Bf4
<HT>
(2) 7-> 5.45 0.29 1. d4 Nf6 2. Nc3 d5 3. Nf3 Bf5 4. Bf4
<HT>
Different scores and moves and PVs at the root...
>
>Nevertheless *no one* is using minimax without alfabeta pruning,
>simply because alfabeta is using less nodes to get the same result.
>
>Alfabeta *might* prune something which minimax doesn't, because of
>the qsearch where you prune on beta anyway.
Minimax does _no_ pruning of any kind, so I have no idea what you are
talking about.
>
>Same is true for nullmove.
I again have no idea what is true for null-move. I don't see how null-move
and alpha/beta can be mentioned in the same sentence since they have _nothing_
to do with each other.
>
>In reality the difference between using nullmove and not using it
>is many plies. Not a constant number, but it IMPROVES the branching
>factor.
Certainly it does. _and_ it changes the move choice and score for moves
as well... as simply proven above.
>
>Nowadays nearly no one is not using hashtables+nullmove, without
>replacing it by very dubious pruning.
>
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.