Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Verified Null-Move Pruning, ICGA 25(3)

Author: Vincent Diepeveen

Date: 19:56:40 11/21/02

Go up one level in this thread


On November 21, 2002 at 21:39:06, Uri Blass wrote:



>On November 21, 2002 at 21:21:09, Vincent Diepeveen wrote:
>
>>On November 20, 2002 at 17:51:40, Alessandro Damiani wrote:
>>
>>'verified' nullmove, or in a different implementation but
>>same algoritm, with just 1 ply reduction is nearly a fullwidth
>>search.
>>
>>I did with a bigger reduction of course. that's also very
>>costly compared to R=3. This was just an experiment carried
>>out years ago when it was described in ICCA journal. Now
>>we have same algoritm in a few lines diff algorithm.
>>
>>I do not see how Omid can just suffer 50% slowdown of his
>>algorithm. Note he only publishes search depths not
>>search times. That is wrong.
>>
>>You must publish search times.
>
>I do not see how you can suffer more than 50% slow down.

Of course you suffer more than 50% slowdown. Every person with
a decent chessprogram will suffer more than 50% slowdown.

This is trivial.

>I think that you simply do not understand the algorithm.

I understand it perfectly well. I have run with it for years.

So please apologize to me!

>The algorithm does not do nearly full width search because after the first
>reduction the search is normal null move pruning without verification.

Yes it nearly does, it is a reduction of just 1 ply!

With nullmove you reduce in the same subtree 4 ply. Or with R=2 you
reduce 3 ply!

Do you understand that with a branching factor of 3.5 that
a search depth of 2 ply more means about a factor of 10 times more
nodes more or less and not factor 2?

>It did not work for movei or for sjeng but the difference is not as big as you
>describe and I suspect that with some modification it may work.
>
>Uri
>>
>>I suffer plies if i reduce just 1 ply.
>>
>>R=2 or R=3 is irrelevant to that.
>>
>>
>>
>>>>
>>>>One final remark: You use standard R = 3 in DIEP. So the search tree constructed
>>>>by your program will definitely be smaller than that of verified R = 3. Many
>>>>people find standard R = 3 as too risky; but if you are happy with its overall
>>>>tactical strength, then I don't recommend you to shift to another method. But
>>>>for those who'd like to get greater tactical strength than standard R = 2, and a
>>>>smaller search tree than R = 2, I recommend to try verified null-move pruning.
>>>>
>>>>Best,
>>>>
>>>>Omid.
>>>>
>>>
>>>Vincent uses R = 3 and complex quiescence search (Vincent, correct me if I am
>>>wrong). Maybe your Verified Null-Move gives about the same results like R = 3
>>>with a complex quiescence search.
>>>
>>>_If_ this is true then your approach is simpler and therefore better. Just my
>>>two thoughts before going to bed. Good nights.......
>>>
>>>Alessandro



This page took 0.02 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.