Author: Vincent Diepeveen
Date: 13:08:34 07/05/01
Go up one level in this thread
On July 05, 2001 at 08:30:20, martin fierz wrote:
>On July 05, 2001 at 07:47:43, Vincent Diepeveen wrote:
>
>>On July 04, 2001 at 07:01:55, martin fierz wrote:
>>
>>>On July 04, 2001 at 05:41:22, Dan Andersson wrote:
>>>
>>>>ProbCut and MultiProbCut:
>>>>http://www.neci.nec.com/homepages/mic/publications.html
>>>>AFAIK this is the canonical source.
>>>
>>>thanks! i had missed the multiprobcut paper there.
>>>
>>>has anyone ever tried this in chess?
>>>
>>>cheers
>>> martin
>>
>>Yes it didn't work here. Apart from that it gametree technical
>>seen is dubious to do, it's reduction factor is too big.
>>
>>Both for my draughts program as well for my chessprogram it didn't work.
>>
>>Fullwidth search is way better for draughts. I get 40 to 60 ply
>>fullwidth in endgame anyway (and still it cannot win won endgames).
>>
>>Best regards,
>>Vincent
>
>hi vincent,
>
>what do you mean by "it's reduction factor is too big"? you can make probcut
>more or less aggressive by changing the probe depth and by changing the cutoff
>value. obviously some settings will make the program play much worse, while
>sensible settings make it play better. at least for my checkers program that is
>true - and it's not just a small improvement... (works for othello too, which is
>a much more complex game than checkers).
>i assumed that the problem of probcut with chess is that there is a king, which
>makes the game hard to predict. for instance, you can sacrifice a large amount
>of material for a mating attack, and the pruning algorithm isn't allowed to
>prune there, else you don't see it.
>in checkers, things are very different, because there is no king. it is usually
>very stupid to sacrifice material. therefore forward pruning gets good results.
>i'm surprised (and i can hardly believe!) that it doesnt work in 10x10 draughts!
>do you really use no pruning at all? can kings in draughts move like queens in
>chess or like kings?
checkers is a simple form of draughts, so the rules are identical with
a few exceptions
a) board in draughts is 10x10, checkers 8x8
b) capturing backwards is allowed in draughts
c) capturing the longest string is forced in draughts in
checkers any capture
d) if you capture over the promotionfield, then in checkers you
change during the capture sequence into a queen, whereas in
draughts you only get a queen if the capture sequence ends
at 10th row.
That's all the diff. further the rules are completely the same.
What i mean is that from a bad program, *any* modification is
seemingly going to work.
I do not know search depths in checkers, but in draughts i already
am in EGTBs at openingsposition, i get in 60 in 1 games around
14 ply or so at tournament level (excluding loads of extensions),
and at that depth a better evaluation helps more as dubious
search enhancements.
pruning on alpha is highly dubious, because you throw the child away
with the bathing water.
It gets backtracked into the tree by hashtables which causes weird effects.
this is not so hard to imagine!
When pieces get off, so does my search depth get up in draughts.
Quickly i go to 40 to 60 ply in far endgame.
Of course the only delaying thing is when you get into EGTBs.
Handling egtbs is far more important there as it is in chess.
>cheers
> martin
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.