Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How important is a root search function ?

Author: scott farrell

Date: 03:47:50 09/05/03

Go up one level in this thread


On September 02, 2003 at 14:54:25, Geoff wrote:

>Hi
>
>Thanks Scott & Richard for the replies, I forgot to mention in my post that I
>had already implemented killer moves, that did help quite a bit as you say.
>
>Out of interest here is my output for nodes and time for the start position.
>What would the concensus be on these figures, good, bad or average ?
>I am guessing my nps is good but my nodes visited is fairly bad compared to
>typical engines.
>I think I will concentrate next on trying to improve move ordering some more
>rather than trying futility pruning and other risky sounding stuff. Looks like I
>had better have a go at a root search function, but I am not that clear yet on
>how to get a better move order at the root.
>
>No hash Table
>Ply  Eval   Time      Nodes   Principle Variation
> 1:    48      0         26   d2d4
> 2:     0      0         97   d2d4 d7d5
> 3:    35      0        635   d2d4 d7d5 g1f3
> 4:     0      0       2218   d2d4 d7d5 g1f3 b8c6
> 5:    35      1      15040   d2d4 d7d5 g1f3 g8f6 b1c3
> 6:    13      6      69361   e2e4 b8c6 d2d4 e7e5 g1f3 g8f6
> 7:    38     32     352671   e2e4 e7e6 b1c3 d7d5 d2d4 g8f6 f1b5
> 8:    18     65     687449   e2e4 e7e6 b1c3 b8c6 g1f3 d7d5 d2d4 g8f6
> 9:    35    282    3064714   e2e4 e7e6 d2d4 b8c6 g1f3 d7d5 e4d5 e6d5 b1c3
>10:    10    804  8,323,244   e2e4 e7e5 g1f3 g8f6 b1c3 b8c6 f1b5 f8b4 d2d3 b4c3
>b2c3
>
>8Meg Hash Table
>Ply  Eval   Time      Nodes   Principle Variation
> 1:    48      0         26   d2d4
> 2:     0      0         97   d2d4 d7d5
> 3:    35      0        632   d2d4 d7d5 g1f3
> 4:     0      0       2080   d2d4 d7d5 g1f3 b8c6
> 5:    35      1      11463   d2d4 d7d5 g1f3 b8c6 b1c3
> 6:    13      4      44199   e2e4 b8c6 d2d4 d7d5 e4d5 d8d5
> 7:    38     18     202541   e2e4 e7e6 d2d4 d7d5 b1c3 g8f6 f1b5
> 8:    18     37     380917   e2e4 e7e6 d2d4 d7d5 b1c3 b8c6 g1f3 g8f6
> 9:    35     90     918522   e2e4
>10:    10    295  2,875,200   e2e4 e7e5 g1f3 g8f6 b1c3 b8c6 f1b5 f8b4 d2d3 b4c3
>b2c3
>            Regards Geoff

here are my numbers:

->1/0(1)  1.99  0.02secs  22  E3
->2/1(3)  0.15  0.1secs  118  E3 E6
->3/3(4)  1.2  0.481secs  391  E3 E6 H4
->4/5(5)  0.2  0.751secs  2165  D4 D5 E3 NF6
->5/6(7)  1.18  0.931secs  5952  D4 D5 NF3 QD6 QD3
->6/7(9)  0.17  2.053secs  44840  E3 E6 D4 NC6 H4 QF6
->7/10(9)  0.73  5.208secs  160404  NC3 E5 E4 NC6 H4 QF6 QF3
d8/11(15) 0/20  -0.15  7.761secs  268557  NC3 E5 E4 NC6 H4 H5 ND5 ND4
d8/11(15) 1/20  -0.05  9.644secs  339436  D4 D5 NF3 NC6 QD3 BG4 BG5
d8/11(15) 3/20  -0.04  12.278secs  452452  E4 D5
->8/11(15)  -0.04  14.301secs  536078  E4 D5
d9/15(18) 0/20  0.38  23.244secs  946895  E4 D5
d9/15(18) 1/20  0.45  48.271secs  1904341  D4 D5 BF4 NF6 NC3 BG4 F3
->9/15(18)  0.45  55.482secs  2160219  D4 D5 BF4 NF6 NC3 BG4 F3
d10/16(20) 0/20  -0.07  96.343secs  3840142  D4 D5 BF4 NF6 NC3 BG4 F3
d10/16(20) 1/20  0.26  119.617secs  4877161  E4 D5 PxD5 QxD5
->10/16(20)  0.26  162.04secs  6690723  E4 D5 PxD5 QxD5


I think it is as much to do with the amount of extensions / what you put in the
qsearch as the actual ordering, and a lot with your eval function. I have heavy
extensions, and therefore many more nodes than you.

notice how my root move ordering is pretty good. If it goes to another move, its
always nearly next.

How I do root move ordering:
1. ply 1-2 each move is searched at -INF,+INF and NOT alpha,beta -
2. plies 2,3,4 -   alpha-QUEEN,beta   alpha-rook,beta alpha-knight,beta
alpha-pawn, beta    respectively , and then take the value, and keep a seperate
ordered table for root based on value. I have some maths to order moves that a
value from a larger ply to eight higher even if the score is slightly lower, as
you start to not get values for all moves at higher plies.
3. normal alpha/beta for root moves after ply 4/5, but still fiddle the ordering
based on fail low/high moves.



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.