Author: Christophe Theron
Date: 22:22:56 05/04/01
Go up one level in this thread
On May 03, 2001 at 18:31:24, Tony Werten wrote:
>On May 03, 2001 at 14:47:17, Christophe Theron wrote:
>
>>On May 03, 2001 at 02:44:49, Ulrich Tuerke wrote:
>>
>>>>...
>>>>>
>>>>>Quiet moves can be interesting too. Giving check can be a waste of time.
>>>>>After all, many are using the node count scheme.
>>>>>So I guess, it can't be bad.
>>>>>
>>>>>I do something different (estimates of scores). It's probably not better.
>>>>>
>>>>>Uli
>>>>
>>>>
>>>>
>>>>I do the same Uli. I see we are at least two in the club. :)
>>>
>>>I'm sure that the club isn't that small. My feeling is that the node counting
>>>isn't bad for root move ordering but there may be even something better.
>>
>>
>>Actually the idea to order root moves by node counter seems very strange to me.
>>
>
>I'm suprised.
>
>If a certain move leads to positions with a lot of extensions, it seems to be
>interesting and it will have a lot of nodes.
>
>If a move leads to a lot cutoffs, it seems bad and will have few nodes.
>
>I found the idea quite logical, it seems to work for my engine and I would have
>guessed it would work even better for more speculative engines like Tiger and
>the King.
>
>cheers,
>
>Tony
I have not tried it, so I should not say anything about it.
However if a move leads to lines that are heavily extended, it does not mean
much for me.
You have to realize that most extensions are pretty useless. Which does not mean
that extending is a useless idea. There is a small percentage of extended lines
that end up having an impact on the main lines, and this small percentage is
enough to make extensions useful.
But to get this small percentage of useful extensions, we unfortunately have to
search a lot of useless extended lines.
The same applies to lines pruned quickly. There might be an error behind the
early pruning of a line, so it does not say much either.
I think the right idea behind the "counting nodes to order" approach is the
number of moves to try before you get a refutation (alphabeta cutoff). This one
I can understand.
But I still do not like it much, because it assumes an intelligent move ordering
at one ply from the root. And most programs are using very simple heuristics for
move ordering after the root, and these heuristics are easily defeated when a
"surprise" arises. And when it happens, the move ordering at one ply from the
root simply looks like random order. So it takes a random node count to find the
other right refutation.
So to me it sounds like you would order moves at the root based on how bad your
move ordering is at one ply from the root...
But I admit that I should first try this idea to look at how it works in
practice.
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.