Author: Christophe Theron
Date: 09:28:44 10/08/99
Go up one level in this thread
On October 08, 1999 at 08:23:14, Robert Hyatt wrote:
>On October 08, 1999 at 03:40:04, Christophe Theron wrote:
>
>>On October 08, 1999 at 00:55:04, Robert Hyatt wrote:
>>
>>>On October 07, 1999 at 23:58:44, Christophe Theron wrote:
>>>
>>>>On October 07, 1999 at 22:55:42, Dave Gomboc wrote:
>>>>
>>>>>On October 07, 1999 at 00:58:07, Christophe Theron wrote:
>>>>>
>>>>>>On October 06, 1999 at 21:52:28, Nicolas Carrasco wrote:
>>>>>>
>>>>>>>Now I am examining how my Alpha Beta cut off's are working!
>>>>>>>
>>>>>>>I would be extreamly pleased to know how many nodes your chess engine make at
>>>>>>>the initial position (if possible without move ordering) at the following
>>>>>>>depths:
>>>>>>
>>>>>>For Chess Tiger 12.0 (with move ordering and selection) the results are:
>>>>>>
>>>>>>2- 46
>>>>>>3- 606
>>>>>>4- 2088
>>>>>>5- 3929
>>>>>>6- 10102
>>>>>>7- 31441
>>>>>>
>>>>>>
>>>>>> Christophe
>>>>>
>>>>>Your odd/even effect is quite a bit larger than I'm used to seeing. Asymmetric
>>>>>pruning?
>>>>>
>>>>>Dave
>>>>
>>>>The odd/even effect depends on the position. For example going from level 2 to
>>>>level 3 takes a lot of work, apparently because the first sequence of capture
>>>>begins, or maybe checks (?). At that time the move ordering stored from the
>>>>previous level becomes completely wrong apparently.
>>>>
>>>>If you look at the increasing (cumulative) branching factor you get:
>>>>
>>>>2 to 3: 14.0x
>>>>3 to 4: 3.4x
>>>>4 to 5: 1.9x
>>>>5 to 6: 2.6x
>>>>6 to 7: 3.1x
>>>>
>>>>So this is not an odd/even effect. Sometimes the odd to even is high, but
>>>>sometimes the even to odd is higher.
>>>>
>>>>As I said, the behaviour is different in each position. Also, at deeper depth it
>>>>tends to be smaller.
>>>>
>>>>
>>>> Christophe
>>>
>>>
>>>Your numbers looked ok to me. Normally even ply searches are easier than odd
>>>ply searches due to alpha beta. IE going from 9 to 10 typically only does a
>>>little more than double the size of the tree, but going from 10 to 11 is a
>>>big change. Because every move at the root is always looked at. And except
>>>for the first root move, at the second, fourth, ..., nth ply (n even) you only
>>>look at one move before exiting with a fail high. For odd depths the last ply
>>>has to examine all moves, and not just 1 move like in even ply searches.
>>>
>>>of course, with extensions, this isn't quite so exact any longer, and it is
>>>possible that they become 'equal'.
>>
>>
>>Yes, that's what I notice generally.
>>
>>
>> Christophe
>
>
>This is a good reason for using a different position. Everyone always wants to
>know "how big is your tree after playing 1. a3?" The question ought to be "how
>big is your tree for the following middlegame position?" Or at least a position
>a few moves into the game. From the original position, hardly any extensions
>fire, so that it is very difficult to compare anything.
When you think about it, the initial position is indeed probably one of the
worse you can find. Captures don't happen in the first plies (or very little of
them), the same for checks.
Also the number of moves for each side is unusually low for a position where all
the pieces are still there.
This is a very singular position in fact.
Christophe
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.