Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question about Alpha-Beta-Improvements

Author: Rafael Andrist

Date: 09:50:21 04/23/01

Go up one level in this thread


On April 23, 2001 at 12:15:09, Antonio Dieguez wrote:

>On April 23, 2001 at 12:04:05, Rafael Andrist wrote:
>
>>On April 23, 2001 at 11:43:33, Antonio Dieguez wrote:
>>
>>>A branching factor around 9 is too high for an alphabeta-search even without
>>>prunning and without hashtable. Why are you using infinite window? try using
>>>small and null windows and see what happen first.
>>>Anyway you say "The use of Iterative Deepening didn't change much" so before how
>>>you calculated the branching factor? The definition I use is nodes iteration
>>>x+1/nodes iteration x, if you are using the other definition that I don't
>>>renember wich is, please forgive my unusefull post.
>>
>>I calculate a virtual branching factor, which is the same for each depth (in
>>reality, it's different).
>>
>>b := branching factor
>>d := depth
>>n := nodes
>>
>>b = n ^ (1 / d)
>>
>>so b ^ d gives n
>>
>>[the ^ means the power function, not the ANSI-Xor]
>>
>>Rafael B. Andrist
>
>This is a weird way because is difficult to compare things isn't? suppose the
>root position has a lot of mobility and possible moves(72)
>
>imagine this:
>
>[1] 100
>[2] 200
>[3] 400
>[4] 800
>[5] 1600
>[6] 3200
>[7] 6400
>
>and a position with a low mobility in the root, imagine this:
>
>[1] 10
>[2] 20
>[3] 40
>[4] 80
>[5] 160
>[6] 320
>[7] 640
>
>using the def I use it's factor 2 in both cases, seems fine. But using yours it
>turns veeeeery weird and different.

You forget that you have a branching factor of 100 at the root is ex. 1 and one
of 10 in ex. 2.

Using "my" definition, it isn't very different:
ex.1: 3.50
ex.2: 2.52

But all the things about mobility are not relevant, because I compare two
identical positions, one searched with and one without iterative deepening.

a practical example from my prog (with bad move sorting):
N: nodes in normal search
Q: nodes in quiescence search
H: number of successful hash access'

with iterative deepening:
1 ply   Sb1-c3  N:       43     Q:       15     H:        0     Value:      5
2 ply   Lf1-b5+ N:      462     Q:      323     H:        9     Value:     -1
3 ply   e4xd5   N:     2695     Q:     3074     H:      106     Value:      2
4 ply   e4-e5   N:    24624     Q:    20999     H:      877     Value:     -5
5 ply   e4xd5   N:   104066     Q:    92941     H:     4173     Value:      1

b = 10.08 (I've calculated it only with the nodes in normal search)

without iterative deepening:
5 ply   e4xd5   N:   148437     Q:   127757     H:     2995     Value:      1

b = 10.82

Rafael B. Andrist



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.